]>
Commit | Line | Data |
---|---|---|
53e1b683 | 1 | /* SPDX-License-Identifier: LGPL-2.1+ */ |
8c6db833 | 2 | |
84ac7bea | 3 | #include <dirent.h> |
8c6db833 | 4 | #include <errno.h> |
84ac7bea | 5 | #include <ftw.h> |
11c3a366 | 6 | #include <limits.h> |
8c6db833 | 7 | #include <signal.h> |
11c3a366 | 8 | #include <stddef.h> |
8c6db833 | 9 | #include <stdlib.h> |
84ac7bea | 10 | #include <string.h> |
672c48cc | 11 | #include <sys/stat.h> |
11c3a366 | 12 | #include <sys/statfs.h> |
672c48cc | 13 | #include <sys/types.h> |
f98c2585 | 14 | #include <sys/utsname.h> |
4b58153d | 15 | #include <sys/xattr.h> |
84ac7bea | 16 | #include <unistd.h> |
8c6db833 | 17 | |
b5efdb8a | 18 | #include "alloc-util.h" |
3ffd4af2 | 19 | #include "cgroup-util.h" |
93cc7779 | 20 | #include "def.h" |
a0956174 | 21 | #include "dirent-util.h" |
84ac7bea | 22 | #include "extract-word.h" |
3ffd4af2 | 23 | #include "fd-util.h" |
84ac7bea | 24 | #include "fileio.h" |
f97b34a6 | 25 | #include "format-util.h" |
f4f15635 | 26 | #include "fs-util.h" |
93cc7779 | 27 | #include "log.h" |
84ac7bea LP |
28 | #include "login-util.h" |
29 | #include "macro.h" | |
93cc7779 | 30 | #include "missing.h" |
84ac7bea | 31 | #include "mkdir.h" |
6bedfcbb | 32 | #include "parse-util.h" |
9eb977db | 33 | #include "path-util.h" |
872a590e | 34 | #include "proc-cmdline.h" |
84ac7bea LP |
35 | #include "process-util.h" |
36 | #include "set.h" | |
9444b1f2 | 37 | #include "special.h" |
872a590e | 38 | #include "stat-util.h" |
d054f0a4 | 39 | #include "stdio-util.h" |
8b43440b | 40 | #include "string-table.h" |
07630cea | 41 | #include "string-util.h" |
aae7e17f | 42 | #include "strv.h" |
84ac7bea | 43 | #include "unit-name.h" |
b1d4f8e1 | 44 | #include "user-util.h" |
8c6db833 | 45 | |
c6c18be3 | 46 | int cg_enumerate_processes(const char *controller, const char *path, FILE **_f) { |
7027ff61 | 47 | _cleanup_free_ char *fs = NULL; |
c6c18be3 | 48 | FILE *f; |
7027ff61 | 49 | int r; |
c6c18be3 | 50 | |
c6c18be3 LP |
51 | assert(_f); |
52 | ||
c3175a7f LP |
53 | r = cg_get_path(controller, path, "cgroup.procs", &fs); |
54 | if (r < 0) | |
c6c18be3 LP |
55 | return r; |
56 | ||
57 | f = fopen(fs, "re"); | |
c6c18be3 LP |
58 | if (!f) |
59 | return -errno; | |
60 | ||
61 | *_f = f; | |
62 | return 0; | |
63 | } | |
64 | ||
c6c18be3 LP |
65 | int cg_read_pid(FILE *f, pid_t *_pid) { |
66 | unsigned long ul; | |
67 | ||
68 | /* Note that the cgroup.procs might contain duplicates! See | |
69 | * cgroups.txt for details. */ | |
70 | ||
7027ff61 LP |
71 | assert(f); |
72 | assert(_pid); | |
73 | ||
c6c18be3 LP |
74 | errno = 0; |
75 | if (fscanf(f, "%lu", &ul) != 1) { | |
76 | ||
77 | if (feof(f)) | |
78 | return 0; | |
79 | ||
f5e5c28f | 80 | return errno > 0 ? -errno : -EIO; |
c6c18be3 LP |
81 | } |
82 | ||
83 | if (ul <= 0) | |
84 | return -EIO; | |
85 | ||
86 | *_pid = (pid_t) ul; | |
87 | return 1; | |
88 | } | |
89 | ||
8b238b13 LP |
90 | int cg_read_event( |
91 | const char *controller, | |
92 | const char *path, | |
93 | const char *event, | |
94 | char **val) { | |
95 | ||
ab2c3861 TH |
96 | _cleanup_free_ char *events = NULL, *content = NULL; |
97 | char *p, *line; | |
98 | int r; | |
99 | ||
100 | r = cg_get_path(controller, path, "cgroup.events", &events); | |
101 | if (r < 0) | |
102 | return r; | |
103 | ||
104 | r = read_full_file(events, &content, NULL); | |
105 | if (r < 0) | |
106 | return r; | |
107 | ||
108 | p = content; | |
109 | while ((line = strsep(&p, "\n"))) { | |
110 | char *key; | |
111 | ||
112 | key = strsep(&line, " "); | |
113 | if (!key || !line) | |
114 | return -EINVAL; | |
115 | ||
116 | if (strcmp(key, event)) | |
117 | continue; | |
118 | ||
119 | *val = strdup(line); | |
120 | return 0; | |
121 | } | |
122 | ||
123 | return -ENOENT; | |
124 | } | |
125 | ||
3228995c CB |
126 | bool cg_ns_supported(void) { |
127 | static thread_local int enabled = -1; | |
128 | ||
129 | if (enabled >= 0) | |
130 | return enabled; | |
131 | ||
0887fa71 LP |
132 | if (access("/proc/self/ns/cgroup", F_OK) < 0) { |
133 | if (errno != ENOENT) | |
134 | log_debug_errno(errno, "Failed to check whether /proc/self/ns/cgroup is available, assuming not: %m"); | |
135 | enabled = false; | |
136 | } else | |
137 | enabled = true; | |
3228995c CB |
138 | |
139 | return enabled; | |
140 | } | |
141 | ||
35d2e7ec | 142 | int cg_enumerate_subgroups(const char *controller, const char *path, DIR **_d) { |
7027ff61 | 143 | _cleanup_free_ char *fs = NULL; |
35d2e7ec LP |
144 | int r; |
145 | DIR *d; | |
146 | ||
35d2e7ec LP |
147 | assert(_d); |
148 | ||
149 | /* This is not recursive! */ | |
150 | ||
c3175a7f LP |
151 | r = cg_get_path(controller, path, NULL, &fs); |
152 | if (r < 0) | |
35d2e7ec LP |
153 | return r; |
154 | ||
155 | d = opendir(fs); | |
35d2e7ec LP |
156 | if (!d) |
157 | return -errno; | |
158 | ||
159 | *_d = d; | |
160 | return 0; | |
161 | } | |
162 | ||
163 | int cg_read_subgroup(DIR *d, char **fn) { | |
164 | struct dirent *de; | |
165 | ||
166 | assert(d); | |
7027ff61 | 167 | assert(fn); |
35d2e7ec | 168 | |
f01327ad | 169 | FOREACH_DIRENT_ALL(de, d, return -errno) { |
35d2e7ec LP |
170 | char *b; |
171 | ||
172 | if (de->d_type != DT_DIR) | |
173 | continue; | |
174 | ||
49bfc877 | 175 | if (dot_or_dot_dot(de->d_name)) |
35d2e7ec LP |
176 | continue; |
177 | ||
7027ff61 LP |
178 | b = strdup(de->d_name); |
179 | if (!b) | |
35d2e7ec LP |
180 | return -ENOMEM; |
181 | ||
182 | *fn = b; | |
183 | return 1; | |
184 | } | |
185 | ||
35d2e7ec LP |
186 | return 0; |
187 | } | |
188 | ||
4ad49000 | 189 | int cg_rmdir(const char *controller, const char *path) { |
7027ff61 | 190 | _cleanup_free_ char *p = NULL; |
35d2e7ec LP |
191 | int r; |
192 | ||
ad293f5a LP |
193 | r = cg_get_path(controller, path, NULL, &p); |
194 | if (r < 0) | |
35d2e7ec LP |
195 | return r; |
196 | ||
197 | r = rmdir(p); | |
7027ff61 LP |
198 | if (r < 0 && errno != ENOENT) |
199 | return -errno; | |
35d2e7ec | 200 | |
b4cccbc1 | 201 | r = cg_hybrid_unified(); |
f20db199 | 202 | if (r <= 0) |
b4cccbc1 | 203 | return r; |
b4cccbc1 LP |
204 | |
205 | if (streq(controller, SYSTEMD_CGROUP_CONTROLLER)) { | |
2977724b TH |
206 | r = cg_rmdir(SYSTEMD_CGROUP_CONTROLLER_LEGACY, path); |
207 | if (r < 0) | |
208 | log_warning_errno(r, "Failed to remove compat systemd cgroup %s: %m", path); | |
209 | } | |
210 | ||
7027ff61 | 211 | return 0; |
35d2e7ec LP |
212 | } |
213 | ||
1d98fef1 LP |
214 | int cg_kill( |
215 | const char *controller, | |
216 | const char *path, | |
217 | int sig, | |
218 | CGroupFlags flags, | |
219 | Set *s, | |
220 | cg_kill_log_func_t log_kill, | |
221 | void *userdata) { | |
222 | ||
7027ff61 | 223 | _cleanup_set_free_ Set *allocated_set = NULL; |
35d2e7ec | 224 | bool done = false; |
c53d2d54 | 225 | int r, ret = 0, ret_log_kill = 0; |
35d2e7ec | 226 | pid_t my_pid; |
8c6db833 | 227 | |
8c6db833 LP |
228 | assert(sig >= 0); |
229 | ||
0d5b4810 LP |
230 | /* Don't send SIGCONT twice. Also, SIGKILL always works even when process is suspended, hence don't send |
231 | * SIGCONT on SIGKILL. */ | |
232 | if (IN_SET(sig, SIGCONT, SIGKILL)) | |
233 | flags &= ~CGROUP_SIGCONT; | |
234 | ||
8c6db833 LP |
235 | /* This goes through the tasks list and kills them all. This |
236 | * is repeated until no further processes are added to the | |
237 | * tasks list, to properly handle forking processes */ | |
238 | ||
7027ff61 | 239 | if (!s) { |
d5099efc | 240 | s = allocated_set = set_new(NULL); |
7027ff61 | 241 | if (!s) |
ca949c9d | 242 | return -ENOMEM; |
7027ff61 | 243 | } |
8c6db833 | 244 | |
df0ff127 | 245 | my_pid = getpid_cached(); |
8c6db833 LP |
246 | |
247 | do { | |
7027ff61 | 248 | _cleanup_fclose_ FILE *f = NULL; |
0b172489 | 249 | pid_t pid = 0; |
8c6db833 LP |
250 | done = true; |
251 | ||
7027ff61 LP |
252 | r = cg_enumerate_processes(controller, path, &f); |
253 | if (r < 0) { | |
4c633005 | 254 | if (ret >= 0 && r != -ENOENT) |
7027ff61 | 255 | return r; |
35d2e7ec | 256 | |
7027ff61 | 257 | return ret; |
35d2e7ec | 258 | } |
c6c18be3 LP |
259 | |
260 | while ((r = cg_read_pid(f, &pid)) > 0) { | |
8c6db833 | 261 | |
1d98fef1 | 262 | if ((flags & CGROUP_IGNORE_SELF) && pid == my_pid) |
c6c18be3 | 263 | continue; |
8c6db833 | 264 | |
fea72cc0 | 265 | if (set_get(s, PID_TO_PTR(pid)) == PID_TO_PTR(pid)) |
c6c18be3 | 266 | continue; |
8c6db833 | 267 | |
1d98fef1 | 268 | if (log_kill) |
c53d2d54 | 269 | ret_log_kill = log_kill(pid, sig, userdata); |
1d98fef1 | 270 | |
8c6db833 LP |
271 | /* If we haven't killed this process yet, kill |
272 | * it */ | |
4c633005 LP |
273 | if (kill(pid, sig) < 0) { |
274 | if (ret >= 0 && errno != ESRCH) | |
8c6db833 | 275 | ret = -errno; |
6e8314c4 | 276 | } else { |
1d98fef1 | 277 | if (flags & CGROUP_SIGCONT) |
e155a0aa | 278 | (void) kill(pid, SIGCONT); |
430c18ed | 279 | |
c53d2d54 DB |
280 | if (ret == 0) { |
281 | if (log_kill) | |
282 | ret = ret_log_kill; | |
283 | else | |
284 | ret = 1; | |
285 | } | |
430c18ed | 286 | } |
8c6db833 | 287 | |
8c6db833 LP |
288 | done = false; |
289 | ||
fea72cc0 | 290 | r = set_put(s, PID_TO_PTR(pid)); |
7027ff61 | 291 | if (r < 0) { |
35d2e7ec | 292 | if (ret >= 0) |
7027ff61 | 293 | return r; |
35d2e7ec | 294 | |
7027ff61 | 295 | return ret; |
35d2e7ec LP |
296 | } |
297 | } | |
298 | ||
299 | if (r < 0) { | |
300 | if (ret >= 0) | |
7027ff61 | 301 | return r; |
35d2e7ec | 302 | |
7027ff61 | 303 | return ret; |
8c6db833 LP |
304 | } |
305 | ||
8c6db833 LP |
306 | /* To avoid racing against processes which fork |
307 | * quicker than we can kill them we repeat this until | |
308 | * no new pids need to be killed. */ | |
309 | ||
35d2e7ec | 310 | } while (!done); |
8c6db833 | 311 | |
35d2e7ec | 312 | return ret; |
8c6db833 LP |
313 | } |
314 | ||
1d98fef1 LP |
315 | int cg_kill_recursive( |
316 | const char *controller, | |
317 | const char *path, | |
318 | int sig, | |
319 | CGroupFlags flags, | |
320 | Set *s, | |
321 | cg_kill_log_func_t log_kill, | |
322 | void *userdata) { | |
323 | ||
7027ff61 LP |
324 | _cleanup_set_free_ Set *allocated_set = NULL; |
325 | _cleanup_closedir_ DIR *d = NULL; | |
e155a0aa | 326 | int r, ret; |
35d2e7ec | 327 | char *fn; |
8c6db833 LP |
328 | |
329 | assert(path); | |
8c6db833 LP |
330 | assert(sig >= 0); |
331 | ||
7027ff61 | 332 | if (!s) { |
d5099efc | 333 | s = allocated_set = set_new(NULL); |
7027ff61 | 334 | if (!s) |
ca949c9d | 335 | return -ENOMEM; |
7027ff61 | 336 | } |
ca949c9d | 337 | |
1d98fef1 | 338 | ret = cg_kill(controller, path, sig, flags, s, log_kill, userdata); |
8c6db833 | 339 | |
7027ff61 LP |
340 | r = cg_enumerate_subgroups(controller, path, &d); |
341 | if (r < 0) { | |
4c633005 | 342 | if (ret >= 0 && r != -ENOENT) |
7027ff61 | 343 | return r; |
8c6db833 | 344 | |
7027ff61 | 345 | return ret; |
35d2e7ec | 346 | } |
8c6db833 | 347 | |
35d2e7ec | 348 | while ((r = cg_read_subgroup(d, &fn)) > 0) { |
7027ff61 | 349 | _cleanup_free_ char *p = NULL; |
8c6db833 | 350 | |
605405c6 | 351 | p = strjoin(path, "/", fn); |
35d2e7ec | 352 | free(fn); |
7027ff61 LP |
353 | if (!p) |
354 | return -ENOMEM; | |
8c6db833 | 355 | |
1d98fef1 | 356 | r = cg_kill_recursive(controller, p, sig, flags, s, log_kill, userdata); |
e155a0aa | 357 | if (r != 0 && ret >= 0) |
35d2e7ec | 358 | ret = r; |
8c6db833 | 359 | } |
7027ff61 | 360 | if (ret >= 0 && r < 0) |
35d2e7ec LP |
361 | ret = r; |
362 | ||
1d98fef1 | 363 | if (flags & CGROUP_REMOVE) { |
4ad49000 | 364 | r = cg_rmdir(controller, path); |
4c701096 | 365 | if (r < 0 && ret >= 0 && !IN_SET(r, -ENOENT, -EBUSY)) |
7027ff61 LP |
366 | return r; |
367 | } | |
ca949c9d | 368 | |
8c6db833 LP |
369 | return ret; |
370 | } | |
371 | ||
1d98fef1 LP |
372 | int cg_migrate( |
373 | const char *cfrom, | |
374 | const char *pfrom, | |
375 | const char *cto, | |
376 | const char *pto, | |
377 | CGroupFlags flags) { | |
378 | ||
35d2e7ec | 379 | bool done = false; |
246aa6dd | 380 | _cleanup_set_free_ Set *s = NULL; |
8c6db833 LP |
381 | int r, ret = 0; |
382 | pid_t my_pid; | |
383 | ||
246aa6dd LP |
384 | assert(cfrom); |
385 | assert(pfrom); | |
386 | assert(cto); | |
387 | assert(pto); | |
8c6db833 | 388 | |
d5099efc | 389 | s = set_new(NULL); |
246aa6dd | 390 | if (!s) |
35d2e7ec LP |
391 | return -ENOMEM; |
392 | ||
df0ff127 | 393 | my_pid = getpid_cached(); |
8c6db833 LP |
394 | |
395 | do { | |
7027ff61 | 396 | _cleanup_fclose_ FILE *f = NULL; |
0b172489 | 397 | pid_t pid = 0; |
8c6db833 LP |
398 | done = true; |
399 | ||
b043cd0b | 400 | r = cg_enumerate_processes(cfrom, pfrom, &f); |
246aa6dd | 401 | if (r < 0) { |
4c633005 | 402 | if (ret >= 0 && r != -ENOENT) |
7027ff61 | 403 | return r; |
35d2e7ec | 404 | |
246aa6dd | 405 | return ret; |
35d2e7ec | 406 | } |
c6c18be3 LP |
407 | |
408 | while ((r = cg_read_pid(f, &pid)) > 0) { | |
8c6db833 | 409 | |
35d2e7ec LP |
410 | /* This might do weird stuff if we aren't a |
411 | * single-threaded program. However, we | |
412 | * luckily know we are not */ | |
1d98fef1 | 413 | if ((flags & CGROUP_IGNORE_SELF) && pid == my_pid) |
c6c18be3 | 414 | continue; |
8c6db833 | 415 | |
fea72cc0 | 416 | if (set_get(s, PID_TO_PTR(pid)) == PID_TO_PTR(pid)) |
35d2e7ec LP |
417 | continue; |
418 | ||
9b84c7f9 LP |
419 | /* Ignore kernel threads. Since they can only |
420 | * exist in the root cgroup, we only check for | |
421 | * them there. */ | |
422 | if (cfrom && | |
57ea45e1 | 423 | empty_or_root(pfrom) && |
9b84c7f9 LP |
424 | is_kernel_thread(pid) > 0) |
425 | continue; | |
426 | ||
246aa6dd LP |
427 | r = cg_attach(cto, pto, pid); |
428 | if (r < 0) { | |
4c633005 | 429 | if (ret >= 0 && r != -ESRCH) |
35d2e7ec LP |
430 | ret = r; |
431 | } else if (ret == 0) | |
432 | ret = 1; | |
8c6db833 | 433 | |
8c6db833 | 434 | done = false; |
35d2e7ec | 435 | |
fea72cc0 | 436 | r = set_put(s, PID_TO_PTR(pid)); |
246aa6dd | 437 | if (r < 0) { |
35d2e7ec | 438 | if (ret >= 0) |
7027ff61 | 439 | return r; |
35d2e7ec | 440 | |
246aa6dd | 441 | return ret; |
35d2e7ec LP |
442 | } |
443 | } | |
444 | ||
445 | if (r < 0) { | |
446 | if (ret >= 0) | |
7027ff61 | 447 | return r; |
35d2e7ec | 448 | |
246aa6dd | 449 | return ret; |
8c6db833 | 450 | } |
35d2e7ec | 451 | } while (!done); |
8c6db833 | 452 | |
35d2e7ec | 453 | return ret; |
8c6db833 LP |
454 | } |
455 | ||
4ad49000 LP |
456 | int cg_migrate_recursive( |
457 | const char *cfrom, | |
458 | const char *pfrom, | |
459 | const char *cto, | |
460 | const char *pto, | |
1d98fef1 | 461 | CGroupFlags flags) { |
4ad49000 | 462 | |
246aa6dd | 463 | _cleanup_closedir_ DIR *d = NULL; |
7027ff61 | 464 | int r, ret = 0; |
35d2e7ec | 465 | char *fn; |
8c6db833 | 466 | |
246aa6dd LP |
467 | assert(cfrom); |
468 | assert(pfrom); | |
469 | assert(cto); | |
470 | assert(pto); | |
8c6db833 | 471 | |
1d98fef1 | 472 | ret = cg_migrate(cfrom, pfrom, cto, pto, flags); |
8c6db833 | 473 | |
246aa6dd LP |
474 | r = cg_enumerate_subgroups(cfrom, pfrom, &d); |
475 | if (r < 0) { | |
4c633005 | 476 | if (ret >= 0 && r != -ENOENT) |
7027ff61 LP |
477 | return r; |
478 | ||
246aa6dd | 479 | return ret; |
35d2e7ec LP |
480 | } |
481 | ||
482 | while ((r = cg_read_subgroup(d, &fn)) > 0) { | |
246aa6dd | 483 | _cleanup_free_ char *p = NULL; |
8c6db833 | 484 | |
605405c6 | 485 | p = strjoin(pfrom, "/", fn); |
35d2e7ec | 486 | free(fn); |
e155a0aa LP |
487 | if (!p) |
488 | return -ENOMEM; | |
8c6db833 | 489 | |
1d98fef1 | 490 | r = cg_migrate_recursive(cfrom, p, cto, pto, flags); |
35d2e7ec LP |
491 | if (r != 0 && ret >= 0) |
492 | ret = r; | |
8c6db833 LP |
493 | } |
494 | ||
35d2e7ec LP |
495 | if (r < 0 && ret >= 0) |
496 | ret = r; | |
497 | ||
1d98fef1 | 498 | if (flags & CGROUP_REMOVE) { |
4ad49000 | 499 | r = cg_rmdir(cfrom, pfrom); |
4c701096 | 500 | if (r < 0 && ret >= 0 && !IN_SET(r, -ENOENT, -EBUSY)) |
246aa6dd LP |
501 | return r; |
502 | } | |
8c6db833 LP |
503 | |
504 | return ret; | |
505 | } | |
506 | ||
13b84ec7 LP |
507 | int cg_migrate_recursive_fallback( |
508 | const char *cfrom, | |
509 | const char *pfrom, | |
510 | const char *cto, | |
511 | const char *pto, | |
1d98fef1 | 512 | CGroupFlags flags) { |
13b84ec7 LP |
513 | |
514 | int r; | |
515 | ||
516 | assert(cfrom); | |
517 | assert(pfrom); | |
518 | assert(cto); | |
519 | assert(pto); | |
520 | ||
1d98fef1 | 521 | r = cg_migrate_recursive(cfrom, pfrom, cto, pto, flags); |
13b84ec7 LP |
522 | if (r < 0) { |
523 | char prefix[strlen(pto) + 1]; | |
524 | ||
525 | /* This didn't work? Then let's try all prefixes of the destination */ | |
526 | ||
fecffe5d | 527 | PATH_FOREACH_PREFIX(prefix, pto) { |
e155a0aa LP |
528 | int q; |
529 | ||
1d98fef1 | 530 | q = cg_migrate_recursive(cfrom, pfrom, cto, prefix, flags); |
e155a0aa LP |
531 | if (q >= 0) |
532 | return q; | |
13b84ec7 LP |
533 | } |
534 | } | |
535 | ||
e155a0aa | 536 | return r; |
13b84ec7 LP |
537 | } |
538 | ||
efdb0237 LP |
539 | static const char *controller_to_dirname(const char *controller) { |
540 | const char *e; | |
3474ae3c | 541 | |
7027ff61 LP |
542 | assert(controller); |
543 | ||
efdb0237 LP |
544 | /* Converts a controller name to the directory name below |
545 | * /sys/fs/cgroup/ we want to mount it to. Effectively, this | |
546 | * just cuts off the name= prefixed used for named | |
547 | * hierarchies, if it is specified. */ | |
548 | ||
2977724b | 549 | if (streq(controller, SYSTEMD_CGROUP_CONTROLLER)) { |
b4cccbc1 | 550 | if (cg_hybrid_unified() > 0) |
2977724b TH |
551 | controller = SYSTEMD_CGROUP_CONTROLLER_HYBRID; |
552 | else | |
553 | controller = SYSTEMD_CGROUP_CONTROLLER_LEGACY; | |
554 | } | |
b6629c4b | 555 | |
efdb0237 LP |
556 | e = startswith(controller, "name="); |
557 | if (e) | |
558 | return e; | |
559 | ||
560 | return controller; | |
3474ae3c LP |
561 | } |
562 | ||
569b19d8 LP |
563 | static int join_path_legacy(const char *controller, const char *path, const char *suffix, char **fs) { |
564 | const char *dn; | |
018ef268 | 565 | char *t = NULL; |
3474ae3c | 566 | |
efdb0237 | 567 | assert(fs); |
569b19d8 LP |
568 | assert(controller); |
569 | ||
570 | dn = controller_to_dirname(controller); | |
efdb0237 LP |
571 | |
572 | if (isempty(path) && isempty(suffix)) | |
569b19d8 | 573 | t = strappend("/sys/fs/cgroup/", dn); |
efdb0237 | 574 | else if (isempty(path)) |
605405c6 | 575 | t = strjoin("/sys/fs/cgroup/", dn, "/", suffix); |
efdb0237 | 576 | else if (isempty(suffix)) |
605405c6 | 577 | t = strjoin("/sys/fs/cgroup/", dn, "/", path); |
efdb0237 | 578 | else |
605405c6 | 579 | t = strjoin("/sys/fs/cgroup/", dn, "/", path, "/", suffix); |
efdb0237 LP |
580 | if (!t) |
581 | return -ENOMEM; | |
3474ae3c | 582 | |
efdb0237 LP |
583 | *fs = t; |
584 | return 0; | |
585 | } | |
586 | ||
587 | static int join_path_unified(const char *path, const char *suffix, char **fs) { | |
588 | char *t; | |
589 | ||
590 | assert(fs); | |
591 | ||
592 | if (isempty(path) && isempty(suffix)) | |
593 | t = strdup("/sys/fs/cgroup"); | |
594 | else if (isempty(path)) | |
595 | t = strappend("/sys/fs/cgroup/", suffix); | |
596 | else if (isempty(suffix)) | |
597 | t = strappend("/sys/fs/cgroup/", path); | |
598 | else | |
605405c6 | 599 | t = strjoin("/sys/fs/cgroup/", path, "/", suffix); |
3474ae3c LP |
600 | if (!t) |
601 | return -ENOMEM; | |
602 | ||
efdb0237 | 603 | *fs = t; |
3474ae3c LP |
604 | return 0; |
605 | } | |
606 | ||
8c6db833 | 607 | int cg_get_path(const char *controller, const char *path, const char *suffix, char **fs) { |
415fc41c | 608 | int r; |
8c6db833 | 609 | |
dbd821ac LP |
610 | assert(fs); |
611 | ||
efdb0237 LP |
612 | if (!controller) { |
613 | char *t; | |
614 | ||
569b19d8 LP |
615 | /* If no controller is specified, we return the path |
616 | * *below* the controllers, without any prefix. */ | |
efdb0237 LP |
617 | |
618 | if (!path && !suffix) | |
619 | return -EINVAL; | |
620 | ||
989189ea | 621 | if (!suffix) |
efdb0237 | 622 | t = strdup(path); |
989189ea | 623 | else if (!path) |
efdb0237 LP |
624 | t = strdup(suffix); |
625 | else | |
605405c6 | 626 | t = strjoin(path, "/", suffix); |
efdb0237 LP |
627 | if (!t) |
628 | return -ENOMEM; | |
629 | ||
858d36c1 | 630 | *fs = path_simplify(t, false); |
efdb0237 LP |
631 | return 0; |
632 | } | |
633 | ||
634 | if (!cg_controller_is_valid(controller)) | |
78edb35a LP |
635 | return -EINVAL; |
636 | ||
b4cccbc1 LP |
637 | r = cg_all_unified(); |
638 | if (r < 0) | |
639 | return r; | |
640 | if (r > 0) | |
efdb0237 | 641 | r = join_path_unified(path, suffix, fs); |
569b19d8 LP |
642 | else |
643 | r = join_path_legacy(controller, path, suffix, fs); | |
efdb0237 LP |
644 | if (r < 0) |
645 | return r; | |
7027ff61 | 646 | |
858d36c1 | 647 | path_simplify(*fs, false); |
efdb0237 | 648 | return 0; |
3474ae3c | 649 | } |
dbd821ac | 650 | |
efdb0237 | 651 | static int controller_is_accessible(const char *controller) { |
b4cccbc1 | 652 | int r; |
37099707 | 653 | |
efdb0237 | 654 | assert(controller); |
37099707 | 655 | |
efdb0237 LP |
656 | /* Checks whether a specific controller is accessible, |
657 | * i.e. its hierarchy mounted. In the unified hierarchy all | |
658 | * controllers are considered accessible, except for the named | |
659 | * hierarchies */ | |
b12afc8c | 660 | |
efdb0237 LP |
661 | if (!cg_controller_is_valid(controller)) |
662 | return -EINVAL; | |
663 | ||
b4cccbc1 LP |
664 | r = cg_all_unified(); |
665 | if (r < 0) | |
666 | return r; | |
667 | if (r > 0) { | |
efdb0237 LP |
668 | /* We don't support named hierarchies if we are using |
669 | * the unified hierarchy. */ | |
670 | ||
671 | if (streq(controller, SYSTEMD_CGROUP_CONTROLLER)) | |
672 | return 0; | |
673 | ||
674 | if (startswith(controller, "name=")) | |
675 | return -EOPNOTSUPP; | |
676 | ||
677 | } else { | |
678 | const char *cc, *dn; | |
679 | ||
680 | dn = controller_to_dirname(controller); | |
681 | cc = strjoina("/sys/fs/cgroup/", dn); | |
682 | ||
683 | if (laccess(cc, F_OK) < 0) | |
684 | return -errno; | |
685 | } | |
37099707 LP |
686 | |
687 | return 0; | |
688 | } | |
689 | ||
3474ae3c | 690 | int cg_get_path_and_check(const char *controller, const char *path, const char *suffix, char **fs) { |
37099707 | 691 | int r; |
dbd821ac | 692 | |
efdb0237 | 693 | assert(controller); |
3474ae3c | 694 | assert(fs); |
70132bd0 | 695 | |
efdb0237 LP |
696 | /* Check if the specified controller is actually accessible */ |
697 | r = controller_is_accessible(controller); | |
37099707 LP |
698 | if (r < 0) |
699 | return r; | |
3474ae3c | 700 | |
efdb0237 | 701 | return cg_get_path(controller, path, suffix, fs); |
8c6db833 LP |
702 | } |
703 | ||
e27796a0 | 704 | static int trim_cb(const char *path, const struct stat *sb, int typeflag, struct FTW *ftwbuf) { |
4ad49000 LP |
705 | assert(path); |
706 | assert(sb); | |
707 | assert(ftwbuf); | |
e27796a0 LP |
708 | |
709 | if (typeflag != FTW_DP) | |
710 | return 0; | |
711 | ||
712 | if (ftwbuf->level < 1) | |
713 | return 0; | |
714 | ||
e155a0aa | 715 | (void) rmdir(path); |
e27796a0 LP |
716 | return 0; |
717 | } | |
718 | ||
8c6db833 | 719 | int cg_trim(const char *controller, const char *path, bool delete_root) { |
7027ff61 | 720 | _cleanup_free_ char *fs = NULL; |
2977724b | 721 | int r = 0, q; |
8c6db833 | 722 | |
8c6db833 LP |
723 | assert(path); |
724 | ||
e27796a0 LP |
725 | r = cg_get_path(controller, path, NULL, &fs); |
726 | if (r < 0) | |
8c6db833 LP |
727 | return r; |
728 | ||
e27796a0 | 729 | errno = 0; |
e155a0aa LP |
730 | if (nftw(fs, trim_cb, 64, FTW_DEPTH|FTW_MOUNT|FTW_PHYS) != 0) { |
731 | if (errno == ENOENT) | |
732 | r = 0; | |
b3267152 | 733 | else if (errno > 0) |
e155a0aa LP |
734 | r = -errno; |
735 | else | |
736 | r = -EIO; | |
737 | } | |
e27796a0 LP |
738 | |
739 | if (delete_root) { | |
4ad49000 LP |
740 | if (rmdir(fs) < 0 && errno != ENOENT) |
741 | return -errno; | |
e27796a0 LP |
742 | } |
743 | ||
b4cccbc1 LP |
744 | q = cg_hybrid_unified(); |
745 | if (q < 0) | |
746 | return q; | |
747 | if (q > 0 && streq(controller, SYSTEMD_CGROUP_CONTROLLER)) { | |
2977724b TH |
748 | q = cg_trim(SYSTEMD_CGROUP_CONTROLLER_LEGACY, path, delete_root); |
749 | if (q < 0) | |
750 | log_warning_errno(q, "Failed to trim compat systemd cgroup %s: %m", path); | |
751 | } | |
752 | ||
e27796a0 | 753 | return r; |
8c6db833 LP |
754 | } |
755 | ||
65be7e06 ZJS |
756 | /* Create a cgroup in the hierarchy of controller. |
757 | * Returns 0 if the group already existed, 1 on success, negative otherwise. | |
758 | */ | |
1434ae6f LP |
759 | int cg_create(const char *controller, const char *path) { |
760 | _cleanup_free_ char *fs = NULL; | |
761 | int r; | |
762 | ||
763 | r = cg_get_path_and_check(controller, path, NULL, &fs); | |
764 | if (r < 0) | |
765 | return r; | |
766 | ||
767 | r = mkdir_parents(fs, 0755); | |
768 | if (r < 0) | |
769 | return r; | |
770 | ||
dae8b82e ZJS |
771 | r = mkdir_errno_wrapper(fs, 0755); |
772 | if (r == -EEXIST) | |
773 | return 0; | |
774 | if (r < 0) | |
775 | return r; | |
1434ae6f | 776 | |
b4cccbc1 LP |
777 | r = cg_hybrid_unified(); |
778 | if (r < 0) | |
779 | return r; | |
780 | ||
781 | if (r > 0 && streq(controller, SYSTEMD_CGROUP_CONTROLLER)) { | |
2977724b TH |
782 | r = cg_create(SYSTEMD_CGROUP_CONTROLLER_LEGACY, path); |
783 | if (r < 0) | |
784 | log_warning_errno(r, "Failed to create compat systemd cgroup %s: %m", path); | |
785 | } | |
786 | ||
1434ae6f LP |
787 | return 1; |
788 | } | |
789 | ||
790 | int cg_create_and_attach(const char *controller, const char *path, pid_t pid) { | |
791 | int r, q; | |
792 | ||
793 | assert(pid >= 0); | |
794 | ||
795 | r = cg_create(controller, path); | |
796 | if (r < 0) | |
797 | return r; | |
798 | ||
799 | q = cg_attach(controller, path, pid); | |
800 | if (q < 0) | |
801 | return q; | |
802 | ||
803 | /* This does not remove the cgroup on failure */ | |
804 | return r; | |
805 | } | |
806 | ||
8c6db833 | 807 | int cg_attach(const char *controller, const char *path, pid_t pid) { |
574d5f2d LP |
808 | _cleanup_free_ char *fs = NULL; |
809 | char c[DECIMAL_STR_MAX(pid_t) + 2]; | |
8c6db833 LP |
810 | int r; |
811 | ||
8c6db833 LP |
812 | assert(path); |
813 | assert(pid >= 0); | |
814 | ||
b043cd0b | 815 | r = cg_get_path_and_check(controller, path, "cgroup.procs", &fs); |
3474ae3c | 816 | if (r < 0) |
c6c18be3 | 817 | return r; |
8c6db833 LP |
818 | |
819 | if (pid == 0) | |
df0ff127 | 820 | pid = getpid_cached(); |
8c6db833 | 821 | |
d054f0a4 | 822 | xsprintf(c, PID_FMT "\n", pid); |
8c6db833 | 823 | |
604028de | 824 | r = write_string_file(fs, c, WRITE_STRING_FILE_DISABLE_BUFFER); |
2977724b TH |
825 | if (r < 0) |
826 | return r; | |
827 | ||
b4cccbc1 LP |
828 | r = cg_hybrid_unified(); |
829 | if (r < 0) | |
830 | return r; | |
831 | ||
832 | if (r > 0 && streq(controller, SYSTEMD_CGROUP_CONTROLLER)) { | |
2977724b TH |
833 | r = cg_attach(SYSTEMD_CGROUP_CONTROLLER_LEGACY, path, pid); |
834 | if (r < 0) | |
bd68e99b | 835 | log_warning_errno(r, "Failed to attach "PID_FMT" to compat systemd cgroup %s: %m", pid, path); |
2977724b TH |
836 | } |
837 | ||
838 | return 0; | |
8c6db833 LP |
839 | } |
840 | ||
13b84ec7 LP |
841 | int cg_attach_fallback(const char *controller, const char *path, pid_t pid) { |
842 | int r; | |
843 | ||
844 | assert(controller); | |
845 | assert(path); | |
846 | assert(pid >= 0); | |
847 | ||
848 | r = cg_attach(controller, path, pid); | |
849 | if (r < 0) { | |
850 | char prefix[strlen(path) + 1]; | |
851 | ||
852 | /* This didn't work? Then let's try all prefixes of | |
853 | * the destination */ | |
854 | ||
fecffe5d | 855 | PATH_FOREACH_PREFIX(prefix, path) { |
e155a0aa LP |
856 | int q; |
857 | ||
858 | q = cg_attach(controller, prefix, pid); | |
859 | if (q >= 0) | |
860 | return q; | |
13b84ec7 LP |
861 | } |
862 | } | |
863 | ||
e155a0aa | 864 | return r; |
13b84ec7 LP |
865 | } |
866 | ||
62b9bb26 | 867 | int cg_set_access( |
2d76d14e LP |
868 | const char *controller, |
869 | const char *path, | |
2d76d14e LP |
870 | uid_t uid, |
871 | gid_t gid) { | |
872 | ||
62b9bb26 LP |
873 | struct Attribute { |
874 | const char *name; | |
875 | bool fatal; | |
876 | }; | |
877 | ||
4e1dfa45 | 878 | /* cgroup v1, aka legacy/non-unified */ |
62b9bb26 LP |
879 | static const struct Attribute legacy_attributes[] = { |
880 | { "cgroup.procs", true }, | |
881 | { "tasks", false }, | |
882 | { "cgroup.clone_children", false }, | |
883 | {}, | |
884 | }; | |
885 | ||
4e1dfa45 | 886 | /* cgroup v2, aka unified */ |
62b9bb26 LP |
887 | static const struct Attribute unified_attributes[] = { |
888 | { "cgroup.procs", true }, | |
889 | { "cgroup.subtree_control", true }, | |
890 | { "cgroup.threads", false }, | |
891 | {}, | |
892 | }; | |
893 | ||
894 | static const struct Attribute* const attributes[] = { | |
895 | [false] = legacy_attributes, | |
896 | [true] = unified_attributes, | |
897 | }; | |
974efc46 | 898 | |
40853aa5 | 899 | _cleanup_free_ char *fs = NULL; |
62b9bb26 LP |
900 | const struct Attribute *i; |
901 | int r, unified; | |
8c6db833 | 902 | |
8c6db833 LP |
903 | assert(path); |
904 | ||
62b9bb26 | 905 | if (uid == UID_INVALID && gid == GID_INVALID) |
8d53b453 LP |
906 | return 0; |
907 | ||
62b9bb26 LP |
908 | unified = cg_unified_controller(controller); |
909 | if (unified < 0) | |
910 | return unified; | |
8c6db833 | 911 | |
62b9bb26 LP |
912 | /* Configure access to the cgroup itself */ |
913 | r = cg_get_path(controller, path, NULL, &fs); | |
974efc46 LP |
914 | if (r < 0) |
915 | return r; | |
8c6db833 | 916 | |
62b9bb26 | 917 | r = chmod_and_chown(fs, 0755, uid, gid); |
b4cccbc1 LP |
918 | if (r < 0) |
919 | return r; | |
40853aa5 | 920 | |
62b9bb26 LP |
921 | /* Configure access to the cgroup's attributes */ |
922 | for (i = attributes[unified]; i->name; i++) { | |
40853aa5 | 923 | fs = mfree(fs); |
40853aa5 | 924 | |
62b9bb26 | 925 | r = cg_get_path(controller, path, i->name, &fs); |
40853aa5 LP |
926 | if (r < 0) |
927 | return r; | |
efdb0237 | 928 | |
62b9bb26 LP |
929 | r = chmod_and_chown(fs, 0644, uid, gid); |
930 | if (r < 0) { | |
931 | if (i->fatal) | |
932 | return r; | |
5beac75e | 933 | |
62b9bb26 LP |
934 | log_debug_errno(r, "Failed to set access on cgroup %s, ignoring: %m", fs); |
935 | } | |
936 | } | |
937 | ||
938 | if (streq(controller, SYSTEMD_CGROUP_CONTROLLER)) { | |
939 | r = cg_hybrid_unified(); | |
2977724b | 940 | if (r < 0) |
62b9bb26 LP |
941 | return r; |
942 | if (r > 0) { | |
943 | /* Always propagate access mode from unified to legacy controller */ | |
944 | r = cg_set_access(SYSTEMD_CGROUP_CONTROLLER_LEGACY, path, uid, gid); | |
945 | if (r < 0) | |
946 | log_debug_errno(r, "Failed to set access on compatibility systemd cgroup %s, ignoring: %m", path); | |
947 | } | |
2977724b | 948 | } |
974efc46 | 949 | |
efdb0237 | 950 | return 0; |
8c6db833 LP |
951 | } |
952 | ||
4b58153d LP |
953 | int cg_set_xattr(const char *controller, const char *path, const char *name, const void *value, size_t size, int flags) { |
954 | _cleanup_free_ char *fs = NULL; | |
955 | int r; | |
956 | ||
957 | assert(path); | |
958 | assert(name); | |
959 | assert(value || size <= 0); | |
960 | ||
961 | r = cg_get_path(controller, path, NULL, &fs); | |
962 | if (r < 0) | |
963 | return r; | |
964 | ||
965 | if (setxattr(fs, name, value, size, flags) < 0) | |
966 | return -errno; | |
967 | ||
968 | return 0; | |
969 | } | |
970 | ||
971 | int cg_get_xattr(const char *controller, const char *path, const char *name, void *value, size_t size) { | |
972 | _cleanup_free_ char *fs = NULL; | |
973 | ssize_t n; | |
974 | int r; | |
975 | ||
976 | assert(path); | |
977 | assert(name); | |
978 | ||
979 | r = cg_get_path(controller, path, NULL, &fs); | |
980 | if (r < 0) | |
981 | return r; | |
982 | ||
983 | n = getxattr(fs, name, value, size); | |
984 | if (n < 0) | |
985 | return -errno; | |
986 | ||
987 | return (int) n; | |
988 | } | |
989 | ||
7027ff61 | 990 | int cg_pid_get_path(const char *controller, pid_t pid, char **path) { |
7027ff61 | 991 | _cleanup_fclose_ FILE *f = NULL; |
b6629c4b | 992 | const char *fs, *controller_str; |
d2b39cb6 | 993 | int unified, r; |
efdb0237 | 994 | size_t cs = 0; |
8c6db833 | 995 | |
8c6db833 | 996 | assert(path); |
c6c18be3 | 997 | assert(pid >= 0); |
8c6db833 | 998 | |
5da38d07 TH |
999 | if (controller) { |
1000 | if (!cg_controller_is_valid(controller)) | |
1001 | return -EINVAL; | |
1002 | } else | |
1003 | controller = SYSTEMD_CGROUP_CONTROLLER; | |
1004 | ||
c22800e4 | 1005 | unified = cg_unified_controller(controller); |
b4cccbc1 LP |
1006 | if (unified < 0) |
1007 | return unified; | |
1008 | if (unified == 0) { | |
b6629c4b TH |
1009 | if (streq(controller, SYSTEMD_CGROUP_CONTROLLER)) |
1010 | controller_str = SYSTEMD_CGROUP_CONTROLLER_LEGACY; | |
1011 | else | |
1012 | controller_str = controller; | |
1013 | ||
1014 | cs = strlen(controller_str); | |
1015 | } | |
7027ff61 | 1016 | |
b68fa010 | 1017 | fs = procfs_file_alloca(pid, "cgroup"); |
fdeea3f4 ZJS |
1018 | r = fopen_unlocked(fs, "re", &f); |
1019 | if (r == -ENOENT) | |
1020 | return -ESRCH; | |
1021 | if (r < 0) | |
1022 | return r; | |
35bbbf85 | 1023 | |
d2b39cb6 LP |
1024 | for (;;) { |
1025 | _cleanup_free_ char *line = NULL; | |
efdb0237 | 1026 | char *e, *p; |
c6c18be3 | 1027 | |
d2b39cb6 LP |
1028 | r = read_line(f, LONG_LINE_MAX, &line); |
1029 | if (r < 0) | |
1030 | return r; | |
1031 | if (r == 0) | |
1032 | break; | |
c6c18be3 | 1033 | |
efdb0237 LP |
1034 | if (unified) { |
1035 | e = startswith(line, "0:"); | |
1036 | if (!e) | |
1037 | continue; | |
c6c18be3 | 1038 | |
efdb0237 LP |
1039 | e = strchr(e, ':'); |
1040 | if (!e) | |
1041 | continue; | |
1042 | } else { | |
1043 | char *l; | |
1044 | size_t k; | |
1045 | const char *word, *state; | |
1046 | bool found = false; | |
1047 | ||
1048 | l = strchr(line, ':'); | |
1049 | if (!l) | |
1050 | continue; | |
8af8afd6 | 1051 | |
efdb0237 LP |
1052 | l++; |
1053 | e = strchr(l, ':'); | |
1054 | if (!e) | |
1055 | continue; | |
8af8afd6 | 1056 | |
efdb0237 | 1057 | *e = 0; |
00d4b1e6 | 1058 | FOREACH_WORD_SEPARATOR(word, k, l, ",", state) |
b6629c4b | 1059 | if (k == cs && memcmp(word, controller_str, cs) == 0) { |
efdb0237 LP |
1060 | found = true; |
1061 | break; | |
1062 | } | |
efdb0237 LP |
1063 | if (!found) |
1064 | continue; | |
8af8afd6 LP |
1065 | } |
1066 | ||
8af8afd6 | 1067 | p = strdup(e + 1); |
7027ff61 LP |
1068 | if (!p) |
1069 | return -ENOMEM; | |
c6c18be3 | 1070 | |
5e20b0a4 LP |
1071 | /* Truncate suffix indicating the process is a zombie */ |
1072 | e = endswith(p, " (deleted)"); | |
1073 | if (e) | |
1074 | *e = 0; | |
1075 | ||
c6c18be3 | 1076 | *path = p; |
7027ff61 | 1077 | return 0; |
c6c18be3 LP |
1078 | } |
1079 | ||
1c80e425 | 1080 | return -ENODATA; |
8c6db833 LP |
1081 | } |
1082 | ||
1083 | int cg_install_release_agent(const char *controller, const char *agent) { | |
7027ff61 | 1084 | _cleanup_free_ char *fs = NULL, *contents = NULL; |
efdb0237 | 1085 | const char *sc; |
415fc41c | 1086 | int r; |
8c6db833 | 1087 | |
8c6db833 LP |
1088 | assert(agent); |
1089 | ||
c22800e4 | 1090 | r = cg_unified_controller(controller); |
b4cccbc1 LP |
1091 | if (r < 0) |
1092 | return r; | |
1093 | if (r > 0) /* doesn't apply to unified hierarchy */ | |
efdb0237 LP |
1094 | return -EOPNOTSUPP; |
1095 | ||
7027ff61 LP |
1096 | r = cg_get_path(controller, NULL, "release_agent", &fs); |
1097 | if (r < 0) | |
c6c18be3 | 1098 | return r; |
8c6db833 | 1099 | |
7027ff61 LP |
1100 | r = read_one_line_file(fs, &contents); |
1101 | if (r < 0) | |
1102 | return r; | |
8c6db833 LP |
1103 | |
1104 | sc = strstrip(contents); | |
e155a0aa | 1105 | if (isempty(sc)) { |
604028de | 1106 | r = write_string_file(fs, agent, WRITE_STRING_FILE_DISABLE_BUFFER); |
574d5f2d | 1107 | if (r < 0) |
7027ff61 | 1108 | return r; |
b8725df8 | 1109 | } else if (!path_equal(sc, agent)) |
7027ff61 | 1110 | return -EEXIST; |
8c6db833 | 1111 | |
0da16248 | 1112 | fs = mfree(fs); |
7027ff61 LP |
1113 | r = cg_get_path(controller, NULL, "notify_on_release", &fs); |
1114 | if (r < 0) | |
1115 | return r; | |
8c6db833 | 1116 | |
0da16248 | 1117 | contents = mfree(contents); |
7027ff61 LP |
1118 | r = read_one_line_file(fs, &contents); |
1119 | if (r < 0) | |
1120 | return r; | |
8c6db833 LP |
1121 | |
1122 | sc = strstrip(contents); | |
8c6db833 | 1123 | if (streq(sc, "0")) { |
604028de | 1124 | r = write_string_file(fs, "1", WRITE_STRING_FILE_DISABLE_BUFFER); |
7027ff61 LP |
1125 | if (r < 0) |
1126 | return r; | |
c6c18be3 | 1127 | |
7027ff61 LP |
1128 | return 1; |
1129 | } | |
8c6db833 | 1130 | |
7027ff61 LP |
1131 | if (!streq(sc, "1")) |
1132 | return -EIO; | |
8c6db833 | 1133 | |
7027ff61 | 1134 | return 0; |
8c6db833 LP |
1135 | } |
1136 | ||
ad929bcc KS |
1137 | int cg_uninstall_release_agent(const char *controller) { |
1138 | _cleanup_free_ char *fs = NULL; | |
415fc41c | 1139 | int r; |
efdb0237 | 1140 | |
c22800e4 | 1141 | r = cg_unified_controller(controller); |
b4cccbc1 LP |
1142 | if (r < 0) |
1143 | return r; | |
1144 | if (r > 0) /* Doesn't apply to unified hierarchy */ | |
efdb0237 | 1145 | return -EOPNOTSUPP; |
ad929bcc | 1146 | |
ac9ef333 LP |
1147 | r = cg_get_path(controller, NULL, "notify_on_release", &fs); |
1148 | if (r < 0) | |
1149 | return r; | |
1150 | ||
604028de | 1151 | r = write_string_file(fs, "0", WRITE_STRING_FILE_DISABLE_BUFFER); |
ac9ef333 LP |
1152 | if (r < 0) |
1153 | return r; | |
1154 | ||
0da16248 | 1155 | fs = mfree(fs); |
ac9ef333 | 1156 | |
ad929bcc KS |
1157 | r = cg_get_path(controller, NULL, "release_agent", &fs); |
1158 | if (r < 0) | |
1159 | return r; | |
1160 | ||
604028de | 1161 | r = write_string_file(fs, "", WRITE_STRING_FILE_DISABLE_BUFFER); |
ad929bcc KS |
1162 | if (r < 0) |
1163 | return r; | |
1164 | ||
ac9ef333 | 1165 | return 0; |
ad929bcc KS |
1166 | } |
1167 | ||
6f883237 | 1168 | int cg_is_empty(const char *controller, const char *path) { |
7027ff61 | 1169 | _cleanup_fclose_ FILE *f = NULL; |
efdb0237 | 1170 | pid_t pid; |
7027ff61 | 1171 | int r; |
8c6db833 | 1172 | |
8c6db833 LP |
1173 | assert(path); |
1174 | ||
b043cd0b | 1175 | r = cg_enumerate_processes(controller, path, &f); |
6f883237 | 1176 | if (r == -ENOENT) |
1bcf3fc6 | 1177 | return true; |
c3175a7f | 1178 | if (r < 0) |
6f883237 | 1179 | return r; |
8c6db833 | 1180 | |
6f883237 | 1181 | r = cg_read_pid(f, &pid); |
c6c18be3 LP |
1182 | if (r < 0) |
1183 | return r; | |
8c6db833 | 1184 | |
6f883237 | 1185 | return r == 0; |
8c6db833 LP |
1186 | } |
1187 | ||
6f883237 | 1188 | int cg_is_empty_recursive(const char *controller, const char *path) { |
415fc41c | 1189 | int r; |
8c6db833 | 1190 | |
8c6db833 LP |
1191 | assert(path); |
1192 | ||
6fd66507 | 1193 | /* The root cgroup is always populated */ |
57ea45e1 | 1194 | if (controller && empty_or_root(path)) |
efdb0237 | 1195 | return false; |
6fd66507 | 1196 | |
c22800e4 | 1197 | r = cg_unified_controller(controller); |
b4cccbc1 LP |
1198 | if (r < 0) |
1199 | return r; | |
1200 | if (r > 0) { | |
ab2c3861 | 1201 | _cleanup_free_ char *t = NULL; |
8c6db833 | 1202 | |
efdb0237 | 1203 | /* On the unified hierarchy we can check empty state |
ab2c3861 | 1204 | * via the "populated" attribute of "cgroup.events". */ |
8c6db833 | 1205 | |
ab2c3861 | 1206 | r = cg_read_event(controller, path, "populated", &t); |
1bcf3fc6 ZJS |
1207 | if (r == -ENOENT) |
1208 | return true; | |
efdb0237 LP |
1209 | if (r < 0) |
1210 | return r; | |
1211 | ||
1212 | return streq(t, "0"); | |
1213 | } else { | |
1214 | _cleanup_closedir_ DIR *d = NULL; | |
1215 | char *fn; | |
8c6db833 | 1216 | |
efdb0237 | 1217 | r = cg_is_empty(controller, path); |
35d2e7ec | 1218 | if (r <= 0) |
7027ff61 | 1219 | return r; |
35d2e7ec | 1220 | |
efdb0237 LP |
1221 | r = cg_enumerate_subgroups(controller, path, &d); |
1222 | if (r == -ENOENT) | |
1bcf3fc6 | 1223 | return true; |
efdb0237 LP |
1224 | if (r < 0) |
1225 | return r; | |
35d2e7ec | 1226 | |
efdb0237 LP |
1227 | while ((r = cg_read_subgroup(d, &fn)) > 0) { |
1228 | _cleanup_free_ char *p = NULL; | |
1229 | ||
605405c6 | 1230 | p = strjoin(path, "/", fn); |
efdb0237 LP |
1231 | free(fn); |
1232 | if (!p) | |
1233 | return -ENOMEM; | |
1234 | ||
1235 | r = cg_is_empty_recursive(controller, p); | |
1236 | if (r <= 0) | |
1237 | return r; | |
1238 | } | |
1239 | if (r < 0) | |
1240 | return r; | |
1241 | ||
1242 | return true; | |
1243 | } | |
35d2e7ec LP |
1244 | } |
1245 | ||
1246 | int cg_split_spec(const char *spec, char **controller, char **path) { | |
35d2e7ec | 1247 | char *t = NULL, *u = NULL; |
efdb0237 | 1248 | const char *e; |
35d2e7ec LP |
1249 | |
1250 | assert(spec); | |
35d2e7ec LP |
1251 | |
1252 | if (*spec == '/') { | |
99be45a4 | 1253 | if (!path_is_normalized(spec)) |
e884315e | 1254 | return -EINVAL; |
35d2e7ec LP |
1255 | |
1256 | if (path) { | |
246aa6dd LP |
1257 | t = strdup(spec); |
1258 | if (!t) | |
35d2e7ec LP |
1259 | return -ENOMEM; |
1260 | ||
858d36c1 | 1261 | *path = path_simplify(t, false); |
8c6db833 LP |
1262 | } |
1263 | ||
35d2e7ec LP |
1264 | if (controller) |
1265 | *controller = NULL; | |
1266 | ||
1267 | return 0; | |
8c6db833 LP |
1268 | } |
1269 | ||
246aa6dd LP |
1270 | e = strchr(spec, ':'); |
1271 | if (!e) { | |
185a0874 | 1272 | if (!cg_controller_is_valid(spec)) |
35d2e7ec LP |
1273 | return -EINVAL; |
1274 | ||
1275 | if (controller) { | |
efdb0237 | 1276 | t = strdup(spec); |
246aa6dd | 1277 | if (!t) |
35d2e7ec LP |
1278 | return -ENOMEM; |
1279 | ||
1280 | *controller = t; | |
1281 | } | |
1282 | ||
1283 | if (path) | |
1284 | *path = NULL; | |
1285 | ||
1286 | return 0; | |
8c6db833 LP |
1287 | } |
1288 | ||
efdb0237 | 1289 | t = strndup(spec, e-spec); |
e884315e LP |
1290 | if (!t) |
1291 | return -ENOMEM; | |
185a0874 | 1292 | if (!cg_controller_is_valid(t)) { |
e884315e | 1293 | free(t); |
35d2e7ec | 1294 | return -EINVAL; |
246aa6dd LP |
1295 | } |
1296 | ||
efdb0237 LP |
1297 | if (isempty(e+1)) |
1298 | u = NULL; | |
1299 | else { | |
baa89da4 LP |
1300 | u = strdup(e+1); |
1301 | if (!u) { | |
1302 | free(t); | |
1303 | return -ENOMEM; | |
1304 | } | |
35d2e7ec | 1305 | |
99be45a4 | 1306 | if (!path_is_normalized(u) || |
baa89da4 LP |
1307 | !path_is_absolute(u)) { |
1308 | free(t); | |
1309 | free(u); | |
1310 | return -EINVAL; | |
1311 | } | |
1312 | ||
858d36c1 | 1313 | path_simplify(u, false); |
baa89da4 | 1314 | } |
5954c074 | 1315 | |
35d2e7ec LP |
1316 | if (controller) |
1317 | *controller = t; | |
e884315e LP |
1318 | else |
1319 | free(t); | |
35d2e7ec LP |
1320 | |
1321 | if (path) | |
1322 | *path = u; | |
e884315e LP |
1323 | else |
1324 | free(u); | |
35d2e7ec LP |
1325 | |
1326 | return 0; | |
8c6db833 | 1327 | } |
c6c18be3 | 1328 | |
7027ff61 | 1329 | int cg_mangle_path(const char *path, char **result) { |
78edb35a LP |
1330 | _cleanup_free_ char *c = NULL, *p = NULL; |
1331 | char *t; | |
35d2e7ec LP |
1332 | int r; |
1333 | ||
1334 | assert(path); | |
1335 | assert(result); | |
1336 | ||
73e231ab | 1337 | /* First, check if it already is a filesystem path */ |
7027ff61 | 1338 | if (path_startswith(path, "/sys/fs/cgroup")) { |
35d2e7ec | 1339 | |
b69d29ce LP |
1340 | t = strdup(path); |
1341 | if (!t) | |
35d2e7ec LP |
1342 | return -ENOMEM; |
1343 | ||
858d36c1 | 1344 | *result = path_simplify(t, false); |
35d2e7ec LP |
1345 | return 0; |
1346 | } | |
1347 | ||
73e231ab | 1348 | /* Otherwise, treat it as cg spec */ |
b69d29ce LP |
1349 | r = cg_split_spec(path, &c, &p); |
1350 | if (r < 0) | |
35d2e7ec LP |
1351 | return r; |
1352 | ||
efdb0237 | 1353 | return cg_get_path(c ?: SYSTEMD_CGROUP_CONTROLLER, p ?: "/", NULL, result); |
35d2e7ec | 1354 | } |
1f73f0f1 | 1355 | |
7027ff61 | 1356 | int cg_get_root_path(char **path) { |
9444b1f2 | 1357 | char *p, *e; |
7027ff61 LP |
1358 | int r; |
1359 | ||
1360 | assert(path); | |
1361 | ||
9444b1f2 | 1362 | r = cg_pid_get_path(SYSTEMD_CGROUP_CONTROLLER, 1, &p); |
7027ff61 LP |
1363 | if (r < 0) |
1364 | return r; | |
1365 | ||
efdb0237 LP |
1366 | e = endswith(p, "/" SPECIAL_INIT_SCOPE); |
1367 | if (!e) | |
1368 | e = endswith(p, "/" SPECIAL_SYSTEM_SLICE); /* legacy */ | |
1369 | if (!e) | |
1370 | e = endswith(p, "/system"); /* even more legacy */ | |
9444b1f2 | 1371 | if (e) |
7027ff61 LP |
1372 | *e = 0; |
1373 | ||
1f73f0f1 LP |
1374 | *path = p; |
1375 | return 0; | |
1376 | } | |
b59e2465 | 1377 | |
751bc6ac LP |
1378 | int cg_shift_path(const char *cgroup, const char *root, const char **shifted) { |
1379 | _cleanup_free_ char *rt = NULL; | |
1380 | char *p; | |
ba1261bc LP |
1381 | int r; |
1382 | ||
e9174f29 | 1383 | assert(cgroup); |
751bc6ac | 1384 | assert(shifted); |
e9174f29 LP |
1385 | |
1386 | if (!root) { | |
1387 | /* If the root was specified let's use that, otherwise | |
1388 | * let's determine it from PID 1 */ | |
1389 | ||
751bc6ac | 1390 | r = cg_get_root_path(&rt); |
e9174f29 LP |
1391 | if (r < 0) |
1392 | return r; | |
1393 | ||
751bc6ac | 1394 | root = rt; |
e9174f29 | 1395 | } |
ba1261bc | 1396 | |
751bc6ac | 1397 | p = path_startswith(cgroup, root); |
efdb0237 | 1398 | if (p && p > cgroup) |
751bc6ac LP |
1399 | *shifted = p - 1; |
1400 | else | |
1401 | *shifted = cgroup; | |
1402 | ||
1403 | return 0; | |
1404 | } | |
1405 | ||
1406 | int cg_pid_get_path_shifted(pid_t pid, const char *root, char **cgroup) { | |
1407 | _cleanup_free_ char *raw = NULL; | |
1408 | const char *c; | |
1409 | int r; | |
1410 | ||
1411 | assert(pid >= 0); | |
1412 | assert(cgroup); | |
1413 | ||
1414 | r = cg_pid_get_path(SYSTEMD_CGROUP_CONTROLLER, pid, &raw); | |
7027ff61 | 1415 | if (r < 0) |
ba1261bc | 1416 | return r; |
ba1261bc | 1417 | |
751bc6ac LP |
1418 | r = cg_shift_path(raw, root, &c); |
1419 | if (r < 0) | |
1420 | return r; | |
ba1261bc | 1421 | |
ae2a15bc LP |
1422 | if (c == raw) |
1423 | *cgroup = TAKE_PTR(raw); | |
1424 | else { | |
751bc6ac | 1425 | char *n; |
ba1261bc | 1426 | |
751bc6ac LP |
1427 | n = strdup(c); |
1428 | if (!n) | |
ba1261bc | 1429 | return -ENOMEM; |
ba1261bc | 1430 | |
751bc6ac LP |
1431 | *cgroup = n; |
1432 | } | |
ba1261bc LP |
1433 | |
1434 | return 0; | |
1435 | } | |
1436 | ||
9ed794a3 | 1437 | int cg_path_decode_unit(const char *cgroup, char **unit) { |
8b0849e9 LP |
1438 | char *c, *s; |
1439 | size_t n; | |
ef1673d1 MT |
1440 | |
1441 | assert(cgroup); | |
6c03089c | 1442 | assert(unit); |
ef1673d1 | 1443 | |
8b0849e9 LP |
1444 | n = strcspn(cgroup, "/"); |
1445 | if (n < 3) | |
1446 | return -ENXIO; | |
1447 | ||
1448 | c = strndupa(cgroup, n); | |
ae018d9b | 1449 | c = cg_unescape(c); |
ef1673d1 | 1450 | |
7410616c | 1451 | if (!unit_name_is_valid(c, UNIT_NAME_PLAIN|UNIT_NAME_INSTANCE)) |
cfeaa44a | 1452 | return -ENXIO; |
ef1673d1 | 1453 | |
d7bd3de0 | 1454 | s = strdup(c); |
6c03089c LP |
1455 | if (!s) |
1456 | return -ENOMEM; | |
1457 | ||
1458 | *unit = s; | |
ef1673d1 MT |
1459 | return 0; |
1460 | } | |
1461 | ||
8b0849e9 LP |
1462 | static bool valid_slice_name(const char *p, size_t n) { |
1463 | ||
1464 | if (!p) | |
1465 | return false; | |
1466 | ||
fbd0b64f | 1467 | if (n < STRLEN("x.slice")) |
8b0849e9 LP |
1468 | return false; |
1469 | ||
1470 | if (memcmp(p + n - 6, ".slice", 6) == 0) { | |
1471 | char buf[n+1], *c; | |
1472 | ||
1473 | memcpy(buf, p, n); | |
1474 | buf[n] = 0; | |
1475 | ||
1476 | c = cg_unescape(buf); | |
1477 | ||
7410616c | 1478 | return unit_name_is_valid(c, UNIT_NAME_PLAIN); |
8b0849e9 LP |
1479 | } |
1480 | ||
1481 | return false; | |
1482 | } | |
1483 | ||
9444b1f2 | 1484 | static const char *skip_slices(const char *p) { |
8b0849e9 LP |
1485 | assert(p); |
1486 | ||
9444b1f2 LP |
1487 | /* Skips over all slice assignments */ |
1488 | ||
1489 | for (;;) { | |
1021b21b LP |
1490 | size_t n; |
1491 | ||
9444b1f2 LP |
1492 | p += strspn(p, "/"); |
1493 | ||
1494 | n = strcspn(p, "/"); | |
8b0849e9 | 1495 | if (!valid_slice_name(p, n)) |
9444b1f2 LP |
1496 | return p; |
1497 | ||
1498 | p += n; | |
1499 | } | |
1500 | } | |
1501 | ||
8b0849e9 | 1502 | int cg_path_get_unit(const char *path, char **ret) { |
6c03089c | 1503 | const char *e; |
8b0849e9 LP |
1504 | char *unit; |
1505 | int r; | |
6c03089c LP |
1506 | |
1507 | assert(path); | |
8b0849e9 | 1508 | assert(ret); |
6c03089c | 1509 | |
9444b1f2 | 1510 | e = skip_slices(path); |
6c03089c | 1511 | |
8b0849e9 LP |
1512 | r = cg_path_decode_unit(e, &unit); |
1513 | if (r < 0) | |
1514 | return r; | |
1515 | ||
1516 | /* We skipped over the slices, don't accept any now */ | |
1517 | if (endswith(unit, ".slice")) { | |
1518 | free(unit); | |
1519 | return -ENXIO; | |
1520 | } | |
1521 | ||
1522 | *ret = unit; | |
1523 | return 0; | |
6c03089c LP |
1524 | } |
1525 | ||
1526 | int cg_pid_get_unit(pid_t pid, char **unit) { | |
7fd1b19b | 1527 | _cleanup_free_ char *cgroup = NULL; |
ba1261bc | 1528 | int r; |
ba1261bc | 1529 | |
ef1673d1 MT |
1530 | assert(unit); |
1531 | ||
7027ff61 | 1532 | r = cg_pid_get_path_shifted(pid, NULL, &cgroup); |
ef1673d1 MT |
1533 | if (r < 0) |
1534 | return r; | |
1535 | ||
6c03089c LP |
1536 | return cg_path_get_unit(cgroup, unit); |
1537 | } | |
ef1673d1 | 1538 | |
d4fffc4b ZJS |
1539 | /** |
1540 | * Skip session-*.scope, but require it to be there. | |
1541 | */ | |
9444b1f2 LP |
1542 | static const char *skip_session(const char *p) { |
1543 | size_t n; | |
1544 | ||
8b0849e9 LP |
1545 | if (isempty(p)) |
1546 | return NULL; | |
9444b1f2 LP |
1547 | |
1548 | p += strspn(p, "/"); | |
1549 | ||
1550 | n = strcspn(p, "/"); | |
fbd0b64f | 1551 | if (n < STRLEN("session-x.scope")) |
d4fffc4b ZJS |
1552 | return NULL; |
1553 | ||
8b0849e9 LP |
1554 | if (memcmp(p, "session-", 8) == 0 && memcmp(p + n - 6, ".scope", 6) == 0) { |
1555 | char buf[n - 8 - 6 + 1]; | |
1556 | ||
1557 | memcpy(buf, p + 8, n - 8 - 6); | |
1558 | buf[n - 8 - 6] = 0; | |
d4fffc4b | 1559 | |
8b0849e9 LP |
1560 | /* Note that session scopes never need unescaping, |
1561 | * since they cannot conflict with the kernel's own | |
1562 | * names, hence we don't need to call cg_unescape() | |
1563 | * here. */ | |
1564 | ||
1565 | if (!session_id_valid(buf)) | |
1566 | return false; | |
1567 | ||
1568 | p += n; | |
1569 | p += strspn(p, "/"); | |
1570 | return p; | |
1571 | } | |
1572 | ||
1573 | return NULL; | |
d4fffc4b ZJS |
1574 | } |
1575 | ||
1576 | /** | |
1577 | * Skip user@*.service, but require it to be there. | |
1578 | */ | |
1579 | static const char *skip_user_manager(const char *p) { | |
1580 | size_t n; | |
1581 | ||
8b0849e9 LP |
1582 | if (isempty(p)) |
1583 | return NULL; | |
d4fffc4b ZJS |
1584 | |
1585 | p += strspn(p, "/"); | |
1586 | ||
1587 | n = strcspn(p, "/"); | |
fbd0b64f | 1588 | if (n < STRLEN("user@x.service")) |
6c03089c | 1589 | return NULL; |
ef1673d1 | 1590 | |
8b0849e9 LP |
1591 | if (memcmp(p, "user@", 5) == 0 && memcmp(p + n - 8, ".service", 8) == 0) { |
1592 | char buf[n - 5 - 8 + 1]; | |
9444b1f2 | 1593 | |
8b0849e9 LP |
1594 | memcpy(buf, p + 5, n - 5 - 8); |
1595 | buf[n - 5 - 8] = 0; | |
1596 | ||
1597 | /* Note that user manager services never need unescaping, | |
1598 | * since they cannot conflict with the kernel's own | |
1599 | * names, hence we don't need to call cg_unescape() | |
1600 | * here. */ | |
1601 | ||
1602 | if (parse_uid(buf, NULL) < 0) | |
1603 | return NULL; | |
1604 | ||
1605 | p += n; | |
1606 | p += strspn(p, "/"); | |
1607 | ||
1608 | return p; | |
1609 | } | |
1610 | ||
1611 | return NULL; | |
9444b1f2 LP |
1612 | } |
1613 | ||
329ac4bc | 1614 | static const char *skip_user_prefix(const char *path) { |
d4fffc4b | 1615 | const char *e, *t; |
ef1673d1 | 1616 | |
6c03089c | 1617 | assert(path); |
ba1261bc | 1618 | |
9444b1f2 LP |
1619 | /* Skip slices, if there are any */ |
1620 | e = skip_slices(path); | |
ba1261bc | 1621 | |
329ac4bc | 1622 | /* Skip the user manager, if it's in the path now... */ |
8b0849e9 | 1623 | t = skip_user_manager(e); |
329ac4bc LP |
1624 | if (t) |
1625 | return t; | |
8b0849e9 | 1626 | |
329ac4bc LP |
1627 | /* Alternatively skip the user session if it is in the path... */ |
1628 | return skip_session(e); | |
1629 | } | |
32081481 | 1630 | |
329ac4bc LP |
1631 | int cg_path_get_user_unit(const char *path, char **ret) { |
1632 | const char *t; | |
6c03089c | 1633 | |
329ac4bc LP |
1634 | assert(path); |
1635 | assert(ret); | |
8b0849e9 | 1636 | |
329ac4bc LP |
1637 | t = skip_user_prefix(path); |
1638 | if (!t) | |
8b0849e9 | 1639 | return -ENXIO; |
8b0849e9 | 1640 | |
329ac4bc LP |
1641 | /* And from here on it looks pretty much the same as for a |
1642 | * system unit, hence let's use the same parser from here | |
1643 | * on. */ | |
1644 | return cg_path_get_unit(t, ret); | |
ef1673d1 | 1645 | } |
ba1261bc | 1646 | |
ef1673d1 | 1647 | int cg_pid_get_user_unit(pid_t pid, char **unit) { |
7fd1b19b | 1648 | _cleanup_free_ char *cgroup = NULL; |
6c03089c LP |
1649 | int r; |
1650 | ||
1651 | assert(unit); | |
1652 | ||
7027ff61 | 1653 | r = cg_pid_get_path_shifted(pid, NULL, &cgroup); |
6c03089c LP |
1654 | if (r < 0) |
1655 | return r; | |
1656 | ||
1657 | return cg_path_get_user_unit(cgroup, unit); | |
ba1261bc | 1658 | } |
e884315e | 1659 | |
7027ff61 | 1660 | int cg_path_get_machine_name(const char *path, char **machine) { |
efdb0237 LP |
1661 | _cleanup_free_ char *u = NULL; |
1662 | const char *sl; | |
89f7c846 | 1663 | int r; |
374ec6ab | 1664 | |
89f7c846 LP |
1665 | r = cg_path_get_unit(path, &u); |
1666 | if (r < 0) | |
1667 | return r; | |
7027ff61 | 1668 | |
efdb0237 | 1669 | sl = strjoina("/run/systemd/machines/unit:", u); |
89f7c846 | 1670 | return readlink_malloc(sl, machine); |
7027ff61 LP |
1671 | } |
1672 | ||
1673 | int cg_pid_get_machine_name(pid_t pid, char **machine) { | |
7fd1b19b | 1674 | _cleanup_free_ char *cgroup = NULL; |
7027ff61 LP |
1675 | int r; |
1676 | ||
1677 | assert(machine); | |
1678 | ||
1679 | r = cg_pid_get_path_shifted(pid, NULL, &cgroup); | |
1680 | if (r < 0) | |
1681 | return r; | |
1682 | ||
1683 | return cg_path_get_machine_name(cgroup, machine); | |
1684 | } | |
1685 | ||
1686 | int cg_path_get_session(const char *path, char **session) { | |
8b0849e9 LP |
1687 | _cleanup_free_ char *unit = NULL; |
1688 | char *start, *end; | |
1689 | int r; | |
7027ff61 LP |
1690 | |
1691 | assert(path); | |
7027ff61 | 1692 | |
8b0849e9 LP |
1693 | r = cg_path_get_unit(path, &unit); |
1694 | if (r < 0) | |
1695 | return r; | |
7027ff61 | 1696 | |
8b0849e9 LP |
1697 | start = startswith(unit, "session-"); |
1698 | if (!start) | |
cfeaa44a | 1699 | return -ENXIO; |
8b0849e9 LP |
1700 | end = endswith(start, ".scope"); |
1701 | if (!end) | |
cfeaa44a | 1702 | return -ENXIO; |
8b0849e9 LP |
1703 | |
1704 | *end = 0; | |
1705 | if (!session_id_valid(start)) | |
cfeaa44a | 1706 | return -ENXIO; |
374ec6ab | 1707 | |
af08d2f9 | 1708 | if (session) { |
8b0849e9 | 1709 | char *rr; |
af08d2f9 | 1710 | |
8b0849e9 LP |
1711 | rr = strdup(start); |
1712 | if (!rr) | |
af08d2f9 LP |
1713 | return -ENOMEM; |
1714 | ||
8b0849e9 | 1715 | *session = rr; |
af08d2f9 | 1716 | } |
7027ff61 | 1717 | |
7027ff61 LP |
1718 | return 0; |
1719 | } | |
1720 | ||
1721 | int cg_pid_get_session(pid_t pid, char **session) { | |
7fd1b19b | 1722 | _cleanup_free_ char *cgroup = NULL; |
7027ff61 LP |
1723 | int r; |
1724 | ||
7027ff61 LP |
1725 | r = cg_pid_get_path_shifted(pid, NULL, &cgroup); |
1726 | if (r < 0) | |
1727 | return r; | |
1728 | ||
1729 | return cg_path_get_session(cgroup, session); | |
1730 | } | |
1731 | ||
ae018d9b | 1732 | int cg_path_get_owner_uid(const char *path, uid_t *uid) { |
374ec6ab | 1733 | _cleanup_free_ char *slice = NULL; |
8b0849e9 | 1734 | char *start, *end; |
374ec6ab | 1735 | int r; |
ae018d9b LP |
1736 | |
1737 | assert(path); | |
ae018d9b | 1738 | |
374ec6ab LP |
1739 | r = cg_path_get_slice(path, &slice); |
1740 | if (r < 0) | |
1741 | return r; | |
ae018d9b | 1742 | |
674eb685 LP |
1743 | start = startswith(slice, "user-"); |
1744 | if (!start) | |
cfeaa44a | 1745 | return -ENXIO; |
8b0849e9 | 1746 | end = endswith(start, ".slice"); |
674eb685 | 1747 | if (!end) |
cfeaa44a | 1748 | return -ENXIO; |
ae018d9b | 1749 | |
8b0849e9 LP |
1750 | *end = 0; |
1751 | if (parse_uid(start, uid) < 0) | |
cfeaa44a | 1752 | return -ENXIO; |
674eb685 | 1753 | |
674eb685 | 1754 | return 0; |
ae018d9b LP |
1755 | } |
1756 | ||
1757 | int cg_pid_get_owner_uid(pid_t pid, uid_t *uid) { | |
1758 | _cleanup_free_ char *cgroup = NULL; | |
1759 | int r; | |
1760 | ||
ae018d9b LP |
1761 | r = cg_pid_get_path_shifted(pid, NULL, &cgroup); |
1762 | if (r < 0) | |
1763 | return r; | |
1764 | ||
1765 | return cg_path_get_owner_uid(cgroup, uid); | |
1766 | } | |
1767 | ||
1021b21b LP |
1768 | int cg_path_get_slice(const char *p, char **slice) { |
1769 | const char *e = NULL; | |
1021b21b LP |
1770 | |
1771 | assert(p); | |
1772 | assert(slice); | |
1773 | ||
329ac4bc LP |
1774 | /* Finds the right-most slice unit from the beginning, but |
1775 | * stops before we come to the first non-slice unit. */ | |
1776 | ||
1021b21b LP |
1777 | for (;;) { |
1778 | size_t n; | |
1779 | ||
1780 | p += strspn(p, "/"); | |
1781 | ||
1782 | n = strcspn(p, "/"); | |
8b0849e9 | 1783 | if (!valid_slice_name(p, n)) { |
1021b21b | 1784 | |
8b0849e9 LP |
1785 | if (!e) { |
1786 | char *s; | |
1021b21b | 1787 | |
e5d855d3 | 1788 | s = strdup(SPECIAL_ROOT_SLICE); |
8b0849e9 LP |
1789 | if (!s) |
1790 | return -ENOMEM; | |
1021b21b | 1791 | |
8b0849e9 LP |
1792 | *slice = s; |
1793 | return 0; | |
1794 | } | |
1795 | ||
1796 | return cg_path_decode_unit(e, slice); | |
1021b21b LP |
1797 | } |
1798 | ||
1799 | e = p; | |
1021b21b LP |
1800 | p += n; |
1801 | } | |
1802 | } | |
1803 | ||
1804 | int cg_pid_get_slice(pid_t pid, char **slice) { | |
1805 | _cleanup_free_ char *cgroup = NULL; | |
1806 | int r; | |
1807 | ||
1808 | assert(slice); | |
1809 | ||
1810 | r = cg_pid_get_path_shifted(pid, NULL, &cgroup); | |
1811 | if (r < 0) | |
1812 | return r; | |
1813 | ||
1814 | return cg_path_get_slice(cgroup, slice); | |
1815 | } | |
1816 | ||
329ac4bc LP |
1817 | int cg_path_get_user_slice(const char *p, char **slice) { |
1818 | const char *t; | |
1819 | assert(p); | |
1820 | assert(slice); | |
1821 | ||
1822 | t = skip_user_prefix(p); | |
1823 | if (!t) | |
1824 | return -ENXIO; | |
1825 | ||
1826 | /* And now it looks pretty much the same as for a system | |
1827 | * slice, so let's just use the same parser from here on. */ | |
1828 | return cg_path_get_slice(t, slice); | |
1829 | } | |
1830 | ||
1831 | int cg_pid_get_user_slice(pid_t pid, char **slice) { | |
1832 | _cleanup_free_ char *cgroup = NULL; | |
1833 | int r; | |
1834 | ||
1835 | assert(slice); | |
1836 | ||
1837 | r = cg_pid_get_path_shifted(pid, NULL, &cgroup); | |
1838 | if (r < 0) | |
1839 | return r; | |
1840 | ||
1841 | return cg_path_get_user_slice(cgroup, slice); | |
1842 | } | |
1843 | ||
ae018d9b LP |
1844 | char *cg_escape(const char *p) { |
1845 | bool need_prefix = false; | |
1846 | ||
1847 | /* This implements very minimal escaping for names to be used | |
1848 | * as file names in the cgroup tree: any name which might | |
1849 | * conflict with a kernel name or is prefixed with '_' is | |
1850 | * prefixed with a '_'. That way, when reading cgroup names it | |
1851 | * is sufficient to remove a single prefixing underscore if | |
1852 | * there is one. */ | |
1853 | ||
1854 | /* The return value of this function (unlike cg_unescape()) | |
1855 | * needs free()! */ | |
1856 | ||
4c701096 | 1857 | if (IN_SET(p[0], 0, '_', '.') || |
0cbd293e | 1858 | STR_IN_SET(p, "notify_on_release", "release_agent", "tasks") || |
efdb0237 | 1859 | startswith(p, "cgroup.")) |
ae018d9b LP |
1860 | need_prefix = true; |
1861 | else { | |
1862 | const char *dot; | |
1863 | ||
1864 | dot = strrchr(p, '.'); | |
1865 | if (dot) { | |
efdb0237 LP |
1866 | CGroupController c; |
1867 | size_t l = dot - p; | |
ae018d9b | 1868 | |
efdb0237 LP |
1869 | for (c = 0; c < _CGROUP_CONTROLLER_MAX; c++) { |
1870 | const char *n; | |
1871 | ||
1872 | n = cgroup_controller_to_string(c); | |
ae018d9b | 1873 | |
efdb0237 LP |
1874 | if (l != strlen(n)) |
1875 | continue; | |
ae018d9b | 1876 | |
efdb0237 LP |
1877 | if (memcmp(p, n, l) != 0) |
1878 | continue; | |
1879 | ||
1880 | need_prefix = true; | |
1881 | break; | |
ae018d9b LP |
1882 | } |
1883 | } | |
1884 | } | |
1885 | ||
1886 | if (need_prefix) | |
1887 | return strappend("_", p); | |
efdb0237 LP |
1888 | |
1889 | return strdup(p); | |
ae018d9b LP |
1890 | } |
1891 | ||
1892 | char *cg_unescape(const char *p) { | |
1893 | assert(p); | |
1894 | ||
1895 | /* The return value of this function (unlike cg_escape()) | |
1896 | * doesn't need free()! */ | |
1897 | ||
1898 | if (p[0] == '_') | |
1899 | return (char*) p+1; | |
1900 | ||
1901 | return (char*) p; | |
1902 | } | |
78edb35a LP |
1903 | |
1904 | #define CONTROLLER_VALID \ | |
4b549144 | 1905 | DIGITS LETTERS \ |
78edb35a LP |
1906 | "_" |
1907 | ||
185a0874 | 1908 | bool cg_controller_is_valid(const char *p) { |
78edb35a LP |
1909 | const char *t, *s; |
1910 | ||
1911 | if (!p) | |
1912 | return false; | |
1913 | ||
b6629c4b TH |
1914 | if (streq(p, SYSTEMD_CGROUP_CONTROLLER)) |
1915 | return true; | |
1916 | ||
185a0874 DJL |
1917 | s = startswith(p, "name="); |
1918 | if (s) | |
1919 | p = s; | |
78edb35a | 1920 | |
4c701096 | 1921 | if (IN_SET(*p, 0, '_')) |
78edb35a LP |
1922 | return false; |
1923 | ||
1924 | for (t = p; *t; t++) | |
1925 | if (!strchr(CONTROLLER_VALID, *t)) | |
1926 | return false; | |
1927 | ||
1928 | if (t - p > FILENAME_MAX) | |
1929 | return false; | |
1930 | ||
1931 | return true; | |
1932 | } | |
a016b922 LP |
1933 | |
1934 | int cg_slice_to_path(const char *unit, char **ret) { | |
1935 | _cleanup_free_ char *p = NULL, *s = NULL, *e = NULL; | |
1936 | const char *dash; | |
7410616c | 1937 | int r; |
a016b922 LP |
1938 | |
1939 | assert(unit); | |
1940 | assert(ret); | |
1941 | ||
e5d855d3 | 1942 | if (streq(unit, SPECIAL_ROOT_SLICE)) { |
c96cc582 LP |
1943 | char *x; |
1944 | ||
1945 | x = strdup(""); | |
1946 | if (!x) | |
1947 | return -ENOMEM; | |
1948 | *ret = x; | |
1949 | return 0; | |
1950 | } | |
1951 | ||
7410616c | 1952 | if (!unit_name_is_valid(unit, UNIT_NAME_PLAIN)) |
a016b922 LP |
1953 | return -EINVAL; |
1954 | ||
1955 | if (!endswith(unit, ".slice")) | |
1956 | return -EINVAL; | |
1957 | ||
7410616c LP |
1958 | r = unit_name_to_prefix(unit, &p); |
1959 | if (r < 0) | |
1960 | return r; | |
a016b922 LP |
1961 | |
1962 | dash = strchr(p, '-'); | |
e66e5b61 LP |
1963 | |
1964 | /* Don't allow initial dashes */ | |
1965 | if (dash == p) | |
1966 | return -EINVAL; | |
1967 | ||
a016b922 LP |
1968 | while (dash) { |
1969 | _cleanup_free_ char *escaped = NULL; | |
1970 | char n[dash - p + sizeof(".slice")]; | |
1971 | ||
989290db | 1972 | #if HAS_FEATURE_MEMORY_SANITIZER |
1c56d501 ZJS |
1973 | /* msan doesn't instrument stpncpy, so it thinks |
1974 | * n is later used unitialized: | |
1975 | * https://github.com/google/sanitizers/issues/926 | |
1976 | */ | |
1977 | zero(n); | |
1978 | #endif | |
1979 | ||
e66e5b61 | 1980 | /* Don't allow trailing or double dashes */ |
4c701096 | 1981 | if (IN_SET(dash[1], 0, '-')) |
c96cc582 | 1982 | return -EINVAL; |
a016b922 | 1983 | |
c96cc582 | 1984 | strcpy(stpncpy(n, p, dash - p), ".slice"); |
7410616c | 1985 | if (!unit_name_is_valid(n, UNIT_NAME_PLAIN)) |
a016b922 LP |
1986 | return -EINVAL; |
1987 | ||
1988 | escaped = cg_escape(n); | |
1989 | if (!escaped) | |
1990 | return -ENOMEM; | |
1991 | ||
1992 | if (!strextend(&s, escaped, "/", NULL)) | |
1993 | return -ENOMEM; | |
1994 | ||
1995 | dash = strchr(dash+1, '-'); | |
1996 | } | |
1997 | ||
1998 | e = cg_escape(unit); | |
1999 | if (!e) | |
2000 | return -ENOMEM; | |
2001 | ||
2002 | if (!strextend(&s, e, NULL)) | |
2003 | return -ENOMEM; | |
2004 | ||
ae2a15bc | 2005 | *ret = TAKE_PTR(s); |
a016b922 LP |
2006 | |
2007 | return 0; | |
2008 | } | |
4ad49000 LP |
2009 | |
2010 | int cg_set_attribute(const char *controller, const char *path, const char *attribute, const char *value) { | |
2011 | _cleanup_free_ char *p = NULL; | |
2012 | int r; | |
2013 | ||
2014 | r = cg_get_path(controller, path, attribute, &p); | |
2015 | if (r < 0) | |
2016 | return r; | |
2017 | ||
604028de | 2018 | return write_string_file(p, value, WRITE_STRING_FILE_DISABLE_BUFFER); |
4ad49000 LP |
2019 | } |
2020 | ||
934277fe LP |
2021 | int cg_get_attribute(const char *controller, const char *path, const char *attribute, char **ret) { |
2022 | _cleanup_free_ char *p = NULL; | |
2023 | int r; | |
2024 | ||
2025 | r = cg_get_path(controller, path, attribute, &p); | |
2026 | if (r < 0) | |
2027 | return r; | |
2028 | ||
2029 | return read_one_line_file(p, ret); | |
2030 | } | |
2031 | ||
b734a4ff LP |
2032 | int cg_get_keyed_attribute( |
2033 | const char *controller, | |
2034 | const char *path, | |
2035 | const char *attribute, | |
2036 | char **keys, | |
2037 | char **ret_values) { | |
66ebf6c0 | 2038 | |
b734a4ff | 2039 | _cleanup_free_ char *filename = NULL, *contents = NULL; |
b734a4ff | 2040 | const char *p; |
9177fa9f | 2041 | size_t n, i, n_done = 0; |
b734a4ff LP |
2042 | char **v; |
2043 | int r; | |
2044 | ||
4e1dfa45 | 2045 | /* Reads one or more fields of a cgroup v2 keyed attribute file. The 'keys' parameter should be an strv with |
b734a4ff LP |
2046 | * all keys to retrieve. The 'ret_values' parameter should be passed as string size with the same number of |
2047 | * entries as 'keys'. On success each entry will be set to the value of the matching key. | |
2048 | * | |
2049 | * If the attribute file doesn't exist at all returns ENOENT, if any key is not found returns ENXIO. */ | |
66ebf6c0 TH |
2050 | |
2051 | r = cg_get_path(controller, path, attribute, &filename); | |
2052 | if (r < 0) | |
2053 | return r; | |
2054 | ||
b734a4ff | 2055 | r = read_full_file(filename, &contents, NULL); |
66ebf6c0 TH |
2056 | if (r < 0) |
2057 | return r; | |
2058 | ||
b734a4ff LP |
2059 | n = strv_length(keys); |
2060 | if (n == 0) /* No keys to retrieve? That's easy, we are done then */ | |
2061 | return 0; | |
66ebf6c0 | 2062 | |
b734a4ff LP |
2063 | /* Let's build this up in a temporary array for now in order not to clobber the return parameter on failure */ |
2064 | v = newa0(char*, n); | |
66ebf6c0 | 2065 | |
b734a4ff LP |
2066 | for (p = contents; *p;) { |
2067 | const char *w = NULL; | |
b734a4ff | 2068 | |
9177fa9f ZJS |
2069 | for (i = 0; i < n; i++) |
2070 | if (!v[i]) { | |
b734a4ff LP |
2071 | w = first_word(p, keys[i]); |
2072 | if (w) | |
2073 | break; | |
66ebf6c0 | 2074 | } |
66ebf6c0 | 2075 | |
b734a4ff | 2076 | if (w) { |
b734a4ff LP |
2077 | size_t l; |
2078 | ||
2079 | l = strcspn(w, NEWLINE); | |
9177fa9f ZJS |
2080 | v[i] = strndup(w, l); |
2081 | if (!v[i]) { | |
b734a4ff LP |
2082 | r = -ENOMEM; |
2083 | goto fail; | |
66ebf6c0 | 2084 | } |
b734a4ff | 2085 | |
b734a4ff | 2086 | n_done++; |
b734a4ff LP |
2087 | if (n_done >= n) |
2088 | goto done; | |
2089 | ||
2090 | p = w + l; | |
9177fa9f | 2091 | } else |
b734a4ff | 2092 | p += strcspn(p, NEWLINE); |
b734a4ff LP |
2093 | |
2094 | p += strspn(p, NEWLINE); | |
66ebf6c0 TH |
2095 | } |
2096 | ||
b734a4ff LP |
2097 | r = -ENXIO; |
2098 | ||
2099 | fail: | |
2100 | for (i = 0; i < n; i++) | |
2101 | free(v[i]); | |
2102 | ||
2103 | return r; | |
2104 | ||
2105 | done: | |
2106 | memcpy(ret_values, v, sizeof(char*) * n); | |
66ebf6c0 | 2107 | return 0; |
b734a4ff | 2108 | |
66ebf6c0 TH |
2109 | } |
2110 | ||
efdb0237 LP |
2111 | int cg_create_everywhere(CGroupMask supported, CGroupMask mask, const char *path) { |
2112 | CGroupController c; | |
e353faa0 | 2113 | CGroupMask done; |
65be7e06 | 2114 | bool created; |
415fc41c | 2115 | int r; |
4ad49000 LP |
2116 | |
2117 | /* This one will create a cgroup in our private tree, but also | |
2118 | * duplicate it in the trees specified in mask, and remove it | |
65be7e06 ZJS |
2119 | * in all others. |
2120 | * | |
2121 | * Returns 0 if the group already existed in the systemd hierarchy, | |
2122 | * 1 on success, negative otherwise. | |
2123 | */ | |
4ad49000 LP |
2124 | |
2125 | /* First create the cgroup in our own hierarchy. */ | |
2126 | r = cg_create(SYSTEMD_CGROUP_CONTROLLER, path); | |
2127 | if (r < 0) | |
2128 | return r; | |
490c5a37 | 2129 | created = r; |
4ad49000 | 2130 | |
efdb0237 | 2131 | /* If we are in the unified hierarchy, we are done now */ |
b4cccbc1 LP |
2132 | r = cg_all_unified(); |
2133 | if (r < 0) | |
2134 | return r; | |
2135 | if (r > 0) | |
65be7e06 | 2136 | return created; |
efdb0237 | 2137 | |
e353faa0 LP |
2138 | supported &= CGROUP_MASK_V1; |
2139 | mask = CGROUP_MASK_EXTEND_JOINED(mask); | |
2140 | done = 0; | |
2141 | ||
efdb0237 LP |
2142 | /* Otherwise, do the same in the other hierarchies */ |
2143 | for (c = 0; c < _CGROUP_CONTROLLER_MAX; c++) { | |
2144 | CGroupMask bit = CGROUP_CONTROLLER_TO_MASK(c); | |
2145 | const char *n; | |
2146 | ||
e353faa0 | 2147 | if (!FLAGS_SET(supported, bit)) |
ab275f23 LP |
2148 | continue; |
2149 | ||
e353faa0 LP |
2150 | if (FLAGS_SET(done, bit)) |
2151 | continue; | |
efdb0237 | 2152 | |
e353faa0 | 2153 | n = cgroup_controller_to_string(c); |
f99850a0 | 2154 | if (FLAGS_SET(mask, bit)) |
efdb0237 | 2155 | (void) cg_create(n, path); |
e353faa0 | 2156 | else |
efdb0237 | 2157 | (void) cg_trim(n, path, true); |
e353faa0 LP |
2158 | |
2159 | done |= CGROUP_MASK_EXTEND_JOINED(bit); | |
4ad49000 LP |
2160 | } |
2161 | ||
65be7e06 | 2162 | return created; |
4ad49000 LP |
2163 | } |
2164 | ||
efdb0237 LP |
2165 | int cg_attach_everywhere(CGroupMask supported, const char *path, pid_t pid, cg_migrate_callback_t path_callback, void *userdata) { |
2166 | CGroupController c; | |
e353faa0 | 2167 | CGroupMask done; |
415fc41c | 2168 | int r; |
4ad49000 LP |
2169 | |
2170 | r = cg_attach(SYSTEMD_CGROUP_CONTROLLER, path, pid); | |
13b84ec7 LP |
2171 | if (r < 0) |
2172 | return r; | |
4ad49000 | 2173 | |
b4cccbc1 LP |
2174 | r = cg_all_unified(); |
2175 | if (r < 0) | |
2176 | return r; | |
2177 | if (r > 0) | |
efdb0237 | 2178 | return 0; |
7b3fd631 | 2179 | |
e353faa0 LP |
2180 | supported &= CGROUP_MASK_V1; |
2181 | done = 0; | |
2182 | ||
efdb0237 LP |
2183 | for (c = 0; c < _CGROUP_CONTROLLER_MAX; c++) { |
2184 | CGroupMask bit = CGROUP_CONTROLLER_TO_MASK(c); | |
2185 | const char *p = NULL; | |
7b3fd631 | 2186 | |
e353faa0 | 2187 | if (!FLAGS_SET(supported, bit)) |
ab275f23 LP |
2188 | continue; |
2189 | ||
e353faa0 | 2190 | if (FLAGS_SET(done, bit)) |
efdb0237 | 2191 | continue; |
7b3fd631 | 2192 | |
efdb0237 LP |
2193 | if (path_callback) |
2194 | p = path_callback(bit, userdata); | |
efdb0237 LP |
2195 | if (!p) |
2196 | p = path; | |
4ad49000 | 2197 | |
efdb0237 | 2198 | (void) cg_attach_fallback(cgroup_controller_to_string(c), p, pid); |
e353faa0 | 2199 | done |= CGROUP_MASK_EXTEND_JOINED(bit); |
4ad49000 LP |
2200 | } |
2201 | ||
13b84ec7 | 2202 | return 0; |
4ad49000 LP |
2203 | } |
2204 | ||
efdb0237 | 2205 | int cg_attach_many_everywhere(CGroupMask supported, const char *path, Set* pids, cg_migrate_callback_t path_callback, void *userdata) { |
6c12b52e LP |
2206 | Iterator i; |
2207 | void *pidp; | |
2208 | int r = 0; | |
2209 | ||
2210 | SET_FOREACH(pidp, pids, i) { | |
fea72cc0 | 2211 | pid_t pid = PTR_TO_PID(pidp); |
13b84ec7 | 2212 | int q; |
6c12b52e | 2213 | |
7b3fd631 | 2214 | q = cg_attach_everywhere(supported, path, pid, path_callback, userdata); |
efdb0237 | 2215 | if (q < 0 && r >= 0) |
13b84ec7 | 2216 | r = q; |
6c12b52e LP |
2217 | } |
2218 | ||
2219 | return r; | |
2220 | } | |
2221 | ||
efdb0237 | 2222 | int cg_migrate_everywhere(CGroupMask supported, const char *from, const char *to, cg_migrate_callback_t to_callback, void *userdata) { |
b3c5bad3 | 2223 | CGroupController c; |
e353faa0 | 2224 | CGroupMask done; |
b4cccbc1 | 2225 | int r = 0, q; |
4ad49000 | 2226 | |
13b84ec7 | 2227 | if (!path_equal(from, to)) { |
1d98fef1 | 2228 | r = cg_migrate_recursive(SYSTEMD_CGROUP_CONTROLLER, from, SYSTEMD_CGROUP_CONTROLLER, to, CGROUP_REMOVE); |
13b84ec7 LP |
2229 | if (r < 0) |
2230 | return r; | |
2231 | } | |
4ad49000 | 2232 | |
b4cccbc1 LP |
2233 | q = cg_all_unified(); |
2234 | if (q < 0) | |
2235 | return q; | |
2236 | if (q > 0) | |
efdb0237 | 2237 | return r; |
03b90d4b | 2238 | |
e353faa0 LP |
2239 | supported &= CGROUP_MASK_V1; |
2240 | done = 0; | |
2241 | ||
efdb0237 LP |
2242 | for (c = 0; c < _CGROUP_CONTROLLER_MAX; c++) { |
2243 | CGroupMask bit = CGROUP_CONTROLLER_TO_MASK(c); | |
2244 | const char *p = NULL; | |
03b90d4b | 2245 | |
e353faa0 | 2246 | if (!FLAGS_SET(supported, bit)) |
ab275f23 LP |
2247 | continue; |
2248 | ||
e353faa0 | 2249 | if (FLAGS_SET(done, bit)) |
efdb0237 | 2250 | continue; |
03b90d4b | 2251 | |
efdb0237 LP |
2252 | if (to_callback) |
2253 | p = to_callback(bit, userdata); | |
efdb0237 LP |
2254 | if (!p) |
2255 | p = to; | |
2256 | ||
1d98fef1 | 2257 | (void) cg_migrate_recursive_fallback(SYSTEMD_CGROUP_CONTROLLER, to, cgroup_controller_to_string(c), p, 0); |
e353faa0 | 2258 | done |= CGROUP_MASK_EXTEND_JOINED(bit); |
4ad49000 LP |
2259 | } |
2260 | ||
e353faa0 | 2261 | return r; |
4ad49000 LP |
2262 | } |
2263 | ||
efdb0237 LP |
2264 | int cg_trim_everywhere(CGroupMask supported, const char *path, bool delete_root) { |
2265 | CGroupController c; | |
e353faa0 | 2266 | CGroupMask done; |
b4cccbc1 | 2267 | int r, q; |
4ad49000 LP |
2268 | |
2269 | r = cg_trim(SYSTEMD_CGROUP_CONTROLLER, path, delete_root); | |
2270 | if (r < 0) | |
2271 | return r; | |
2272 | ||
b4cccbc1 LP |
2273 | q = cg_all_unified(); |
2274 | if (q < 0) | |
2275 | return q; | |
2276 | if (q > 0) | |
efdb0237 LP |
2277 | return r; |
2278 | ||
e353faa0 LP |
2279 | supported &= CGROUP_MASK_V1; |
2280 | done = 0; | |
2281 | ||
efdb0237 LP |
2282 | for (c = 0; c < _CGROUP_CONTROLLER_MAX; c++) { |
2283 | CGroupMask bit = CGROUP_CONTROLLER_TO_MASK(c); | |
2284 | ||
e353faa0 | 2285 | if (!FLAGS_SET(supported, bit)) |
ab275f23 LP |
2286 | continue; |
2287 | ||
e353faa0 | 2288 | if (FLAGS_SET(done, bit)) |
efdb0237 | 2289 | continue; |
4ad49000 | 2290 | |
efdb0237 | 2291 | (void) cg_trim(cgroup_controller_to_string(c), path, delete_root); |
e353faa0 | 2292 | done |= CGROUP_MASK_EXTEND_JOINED(bit); |
4ad49000 LP |
2293 | } |
2294 | ||
e353faa0 | 2295 | return r; |
4ad49000 LP |
2296 | } |
2297 | ||
aae7e17f | 2298 | int cg_mask_to_string(CGroupMask mask, char **ret) { |
ec635a2d LP |
2299 | _cleanup_free_ char *s = NULL; |
2300 | size_t n = 0, allocated = 0; | |
2301 | bool space = false; | |
aae7e17f | 2302 | CGroupController c; |
aae7e17f FB |
2303 | |
2304 | assert(ret); | |
2305 | ||
2306 | if (mask == 0) { | |
2307 | *ret = NULL; | |
2308 | return 0; | |
2309 | } | |
2310 | ||
2311 | for (c = 0; c < _CGROUP_CONTROLLER_MAX; c++) { | |
ec635a2d LP |
2312 | const char *k; |
2313 | size_t l; | |
aae7e17f | 2314 | |
f99850a0 | 2315 | if (!FLAGS_SET(mask, CGROUP_CONTROLLER_TO_MASK(c))) |
aae7e17f FB |
2316 | continue; |
2317 | ||
ec635a2d LP |
2318 | k = cgroup_controller_to_string(c); |
2319 | l = strlen(k); | |
2320 | ||
2321 | if (!GREEDY_REALLOC(s, allocated, n + space + l + 1)) | |
2322 | return -ENOMEM; | |
2323 | ||
2324 | if (space) | |
2325 | s[n] = ' '; | |
2326 | memcpy(s + n + space, k, l); | |
2327 | n += space + l; | |
2328 | ||
2329 | space = true; | |
aae7e17f FB |
2330 | } |
2331 | ||
ec635a2d | 2332 | assert(s); |
aae7e17f | 2333 | |
ec635a2d | 2334 | s[n] = 0; |
ae2a15bc | 2335 | *ret = TAKE_PTR(s); |
ec635a2d | 2336 | |
aae7e17f FB |
2337 | return 0; |
2338 | } | |
2339 | ||
38a90d45 LP |
2340 | int cg_mask_from_string(const char *value, CGroupMask *ret) { |
2341 | CGroupMask m = 0; | |
2342 | ||
2343 | assert(ret); | |
aae7e17f FB |
2344 | assert(value); |
2345 | ||
2346 | for (;;) { | |
2347 | _cleanup_free_ char *n = NULL; | |
2348 | CGroupController v; | |
2349 | int r; | |
2350 | ||
2351 | r = extract_first_word(&value, &n, NULL, 0); | |
2352 | if (r < 0) | |
2353 | return r; | |
2354 | if (r == 0) | |
2355 | break; | |
2356 | ||
2357 | v = cgroup_controller_from_string(n); | |
2358 | if (v < 0) | |
2359 | continue; | |
2360 | ||
38a90d45 | 2361 | m |= CGROUP_CONTROLLER_TO_MASK(v); |
aae7e17f | 2362 | } |
38a90d45 LP |
2363 | |
2364 | *ret = m; | |
aae7e17f FB |
2365 | return 0; |
2366 | } | |
2367 | ||
efdb0237 | 2368 | int cg_mask_supported(CGroupMask *ret) { |
38a90d45 | 2369 | CGroupMask mask; |
415fc41c | 2370 | int r; |
efdb0237 | 2371 | |
67558d15 LP |
2372 | /* Determines the mask of supported cgroup controllers. Only includes controllers we can make sense of and that |
2373 | * are actually accessible. Only covers real controllers, i.e. not the CGROUP_CONTROLLER_BPF_xyz | |
2374 | * pseudo-controllers. */ | |
4ad49000 | 2375 | |
b4cccbc1 LP |
2376 | r = cg_all_unified(); |
2377 | if (r < 0) | |
2378 | return r; | |
2379 | if (r > 0) { | |
5f4c5fef | 2380 | _cleanup_free_ char *root = NULL, *controllers = NULL, *path = NULL; |
efdb0237 LP |
2381 | |
2382 | /* In the unified hierarchy we can read the supported | |
2383 | * and accessible controllers from a the top-level | |
2384 | * cgroup attribute */ | |
2385 | ||
5f4c5fef LP |
2386 | r = cg_get_root_path(&root); |
2387 | if (r < 0) | |
2388 | return r; | |
2389 | ||
2390 | r = cg_get_path(SYSTEMD_CGROUP_CONTROLLER, root, "cgroup.controllers", &path); | |
2391 | if (r < 0) | |
2392 | return r; | |
2393 | ||
2394 | r = read_one_line_file(path, &controllers); | |
efdb0237 LP |
2395 | if (r < 0) |
2396 | return r; | |
4ad49000 | 2397 | |
aae7e17f FB |
2398 | r = cg_mask_from_string(controllers, &mask); |
2399 | if (r < 0) | |
2400 | return r; | |
efdb0237 | 2401 | |
03afd780 | 2402 | /* Currently, we support the cpu, memory, io and pids controller in the unified hierarchy, mask |
03a7b521 | 2403 | * everything else off. */ |
03afd780 | 2404 | mask &= CGROUP_MASK_V2; |
efdb0237 LP |
2405 | |
2406 | } else { | |
2407 | CGroupController c; | |
2408 | ||
03afd780 | 2409 | /* In the legacy hierarchy, we check which hierarchies are mounted. */ |
efdb0237 | 2410 | |
38a90d45 | 2411 | mask = 0; |
efdb0237 | 2412 | for (c = 0; c < _CGROUP_CONTROLLER_MAX; c++) { |
03afd780 | 2413 | CGroupMask bit = CGROUP_CONTROLLER_TO_MASK(c); |
efdb0237 LP |
2414 | const char *n; |
2415 | ||
03afd780 LP |
2416 | if (!FLAGS_SET(CGROUP_MASK_V1, bit)) |
2417 | continue; | |
2418 | ||
efdb0237 LP |
2419 | n = cgroup_controller_to_string(c); |
2420 | if (controller_is_accessible(n) >= 0) | |
03afd780 | 2421 | mask |= bit; |
efdb0237 | 2422 | } |
4ad49000 LP |
2423 | } |
2424 | ||
efdb0237 LP |
2425 | *ret = mask; |
2426 | return 0; | |
4ad49000 | 2427 | } |
b12afc8c | 2428 | |
6925a0de LP |
2429 | int cg_kernel_controllers(Set **ret) { |
2430 | _cleanup_set_free_free_ Set *controllers = NULL; | |
b12afc8c | 2431 | _cleanup_fclose_ FILE *f = NULL; |
b12afc8c LP |
2432 | int r; |
2433 | ||
6925a0de | 2434 | assert(ret); |
b12afc8c | 2435 | |
f09e86bc LS |
2436 | /* Determines the full list of kernel-known controllers. Might include controllers we don't actually support |
2437 | * and controllers that aren't currently accessible (because not mounted). This does not include "name=" | |
2438 | * pseudo-controllers. */ | |
e155a0aa | 2439 | |
6925a0de LP |
2440 | controllers = set_new(&string_hash_ops); |
2441 | if (!controllers) | |
2442 | return -ENOMEM; | |
2443 | ||
fdeea3f4 ZJS |
2444 | r = fopen_unlocked("/proc/cgroups", "re", &f); |
2445 | if (r == -ENOENT) { | |
2446 | *ret = NULL; | |
2447 | return 0; | |
b12afc8c | 2448 | } |
fdeea3f4 ZJS |
2449 | if (r < 0) |
2450 | return r; | |
35bbbf85 | 2451 | |
b12afc8c | 2452 | /* Ignore the header line */ |
2351e44d | 2453 | (void) read_line(f, (size_t) -1, NULL); |
b12afc8c LP |
2454 | |
2455 | for (;;) { | |
2456 | char *controller; | |
2457 | int enabled = 0; | |
2458 | ||
2459 | errno = 0; | |
2460 | if (fscanf(f, "%ms %*i %*i %i", &controller, &enabled) != 2) { | |
2461 | ||
2462 | if (feof(f)) | |
2463 | break; | |
2464 | ||
b3267152 | 2465 | if (ferror(f) && errno > 0) |
b12afc8c LP |
2466 | return -errno; |
2467 | ||
2468 | return -EBADMSG; | |
2469 | } | |
2470 | ||
2471 | if (!enabled) { | |
2472 | free(controller); | |
2473 | continue; | |
2474 | } | |
2475 | ||
efdb0237 | 2476 | if (!cg_controller_is_valid(controller)) { |
b12afc8c LP |
2477 | free(controller); |
2478 | return -EBADMSG; | |
2479 | } | |
2480 | ||
2481 | r = set_consume(controllers, controller); | |
2482 | if (r < 0) | |
2483 | return r; | |
2484 | } | |
2485 | ||
1cc6c93a | 2486 | *ret = TAKE_PTR(controllers); |
6925a0de | 2487 | |
b12afc8c LP |
2488 | return 0; |
2489 | } | |
efdb0237 | 2490 | |
5da38d07 TH |
2491 | static thread_local CGroupUnified unified_cache = CGROUP_UNIFIED_UNKNOWN; |
2492 | ||
4e1dfa45 | 2493 | /* The hybrid mode was initially implemented in v232 and simply mounted cgroup2 on /sys/fs/cgroup/systemd. This |
c22800e4 LP |
2494 | * unfortunately broke other tools (such as docker) which expected the v1 "name=systemd" hierarchy on |
2495 | * /sys/fs/cgroup/systemd. From v233 and on, the hybrid mode mountnbs v2 on /sys/fs/cgroup/unified and maintains | |
2496 | * "name=systemd" hierarchy on /sys/fs/cgroup/systemd for compatibility with other tools. | |
f08e9287 | 2497 | * |
c22800e4 LP |
2498 | * To keep live upgrade working, we detect and support v232 layout. When v232 layout is detected, to keep cgroup v2 |
2499 | * process management but disable the compat dual layout, we return %true on | |
2500 | * cg_unified_controller(SYSTEMD_CGROUP_CONTROLLER) and %false on cg_hybrid_unified(). | |
f08e9287 TH |
2501 | */ |
2502 | static thread_local bool unified_systemd_v232; | |
2503 | ||
1fcca10e | 2504 | static int cg_unified_update(void) { |
efdb0237 | 2505 | |
efdb0237 LP |
2506 | struct statfs fs; |
2507 | ||
2508 | /* Checks if we support the unified hierarchy. Returns an | |
2509 | * error when the cgroup hierarchies aren't mounted yet or we | |
2510 | * have any other trouble determining if the unified hierarchy | |
2511 | * is supported. */ | |
2512 | ||
5da38d07 TH |
2513 | if (unified_cache >= CGROUP_UNIFIED_NONE) |
2514 | return 0; | |
efdb0237 LP |
2515 | |
2516 | if (statfs("/sys/fs/cgroup/", &fs) < 0) | |
c028bed1 | 2517 | return log_debug_errno(errno, "statfs(\"/sys/fs/cgroup/\") failed: %m"); |
efdb0237 | 2518 | |
9aa21133 ZJS |
2519 | if (F_TYPE_EQUAL(fs.f_type, CGROUP2_SUPER_MAGIC)) { |
2520 | log_debug("Found cgroup2 on /sys/fs/cgroup/, full unified hierarchy"); | |
5da38d07 | 2521 | unified_cache = CGROUP_UNIFIED_ALL; |
9aa21133 | 2522 | } else if (F_TYPE_EQUAL(fs.f_type, TMPFS_MAGIC)) { |
2977724b | 2523 | if (statfs("/sys/fs/cgroup/unified/", &fs) == 0 && |
f08e9287 | 2524 | F_TYPE_EQUAL(fs.f_type, CGROUP2_SUPER_MAGIC)) { |
9aa21133 | 2525 | log_debug("Found cgroup2 on /sys/fs/cgroup/unified, unified hierarchy for systemd controller"); |
2977724b | 2526 | unified_cache = CGROUP_UNIFIED_SYSTEMD; |
f08e9287 | 2527 | unified_systemd_v232 = false; |
f08e9287 | 2528 | } else { |
2977724b | 2529 | if (statfs("/sys/fs/cgroup/systemd/", &fs) < 0) |
9aa21133 | 2530 | return log_debug_errno(errno, "statfs(\"/sys/fs/cgroup/systemd\" failed: %m"); |
5535d8f7 EV |
2531 | |
2532 | if (F_TYPE_EQUAL(fs.f_type, CGROUP2_SUPER_MAGIC)) { | |
2533 | log_debug("Found cgroup2 on /sys/fs/cgroup/systemd, unified hierarchy for systemd controller (v232 variant)"); | |
2534 | unified_cache = CGROUP_UNIFIED_SYSTEMD; | |
2535 | unified_systemd_v232 = true; | |
2536 | } else if (F_TYPE_EQUAL(fs.f_type, CGROUP_SUPER_MAGIC)) { | |
2537 | log_debug("Found cgroup on /sys/fs/cgroup/systemd, legacy hierarchy"); | |
2538 | unified_cache = CGROUP_UNIFIED_NONE; | |
2539 | } else { | |
2540 | log_debug("Unexpected filesystem type %llx mounted on /sys/fs/cgroup/systemd, assuming legacy hierarchy", | |
9aa21133 | 2541 | (unsigned long long) fs.f_type); |
5535d8f7 | 2542 | unified_cache = CGROUP_UNIFIED_NONE; |
9aa21133 | 2543 | } |
2977724b | 2544 | } |
baaa35ad ZJS |
2545 | } else |
2546 | return log_debug_errno(SYNTHETIC_ERRNO(ENOMEDIUM), | |
2547 | "Unknown filesystem type %llx mounted on /sys/fs/cgroup.", | |
2548 | (unsigned long long)fs.f_type); | |
efdb0237 | 2549 | |
5da38d07 TH |
2550 | return 0; |
2551 | } | |
2552 | ||
c22800e4 | 2553 | int cg_unified_controller(const char *controller) { |
b4cccbc1 | 2554 | int r; |
5da38d07 | 2555 | |
1fcca10e | 2556 | r = cg_unified_update(); |
b4cccbc1 LP |
2557 | if (r < 0) |
2558 | return r; | |
5da38d07 | 2559 | |
fc9ae717 LP |
2560 | if (unified_cache == CGROUP_UNIFIED_NONE) |
2561 | return false; | |
2562 | ||
2563 | if (unified_cache >= CGROUP_UNIFIED_ALL) | |
2564 | return true; | |
2565 | ||
2566 | return streq_ptr(controller, SYSTEMD_CGROUP_CONTROLLER); | |
5da38d07 TH |
2567 | } |
2568 | ||
b4cccbc1 | 2569 | int cg_all_unified(void) { |
4bb652ac LP |
2570 | int r; |
2571 | ||
2572 | r = cg_unified_update(); | |
2573 | if (r < 0) | |
2574 | return r; | |
2575 | ||
2576 | return unified_cache >= CGROUP_UNIFIED_ALL; | |
efdb0237 LP |
2577 | } |
2578 | ||
b4cccbc1 LP |
2579 | int cg_hybrid_unified(void) { |
2580 | int r; | |
2977724b | 2581 | |
1fcca10e | 2582 | r = cg_unified_update(); |
b4cccbc1 LP |
2583 | if (r < 0) |
2584 | return r; | |
2977724b | 2585 | |
f08e9287 | 2586 | return unified_cache == CGROUP_UNIFIED_SYSTEMD && !unified_systemd_v232; |
2977724b TH |
2587 | } |
2588 | ||
415fc41c | 2589 | int cg_unified_flush(void) { |
5da38d07 | 2590 | unified_cache = CGROUP_UNIFIED_UNKNOWN; |
415fc41c | 2591 | |
1fcca10e | 2592 | return cg_unified_update(); |
efdb0237 LP |
2593 | } |
2594 | ||
27adcc97 LP |
2595 | int cg_enable_everywhere( |
2596 | CGroupMask supported, | |
2597 | CGroupMask mask, | |
2598 | const char *p, | |
2599 | CGroupMask *ret_result_mask) { | |
2600 | ||
77fa610b | 2601 | _cleanup_fclose_ FILE *f = NULL; |
efdb0237 LP |
2602 | _cleanup_free_ char *fs = NULL; |
2603 | CGroupController c; | |
27adcc97 | 2604 | CGroupMask ret = 0; |
415fc41c | 2605 | int r; |
efdb0237 LP |
2606 | |
2607 | assert(p); | |
2608 | ||
27adcc97 LP |
2609 | if (supported == 0) { |
2610 | if (ret_result_mask) | |
2611 | *ret_result_mask = 0; | |
efdb0237 | 2612 | return 0; |
27adcc97 | 2613 | } |
efdb0237 | 2614 | |
b4cccbc1 LP |
2615 | r = cg_all_unified(); |
2616 | if (r < 0) | |
2617 | return r; | |
27adcc97 LP |
2618 | if (r == 0) { |
2619 | /* On the legacy hiearchy there's no concept of "enabling" controllers in cgroups defined. Let's claim | |
2620 | * complete success right away. (If you wonder why we return the full mask here, rather than zero: the | |
2621 | * caller tends to use the returned mask later on to compare if all controllers where properly joined, | |
2622 | * and if not requeues realization. This use is the primary purpose of the return value, hence let's | |
2623 | * minimize surprises here and reduce triggers for re-realization by always saying we fully | |
2624 | * succeeded.) */ | |
2625 | if (ret_result_mask) | |
2626 | *ret_result_mask = mask & supported & CGROUP_MASK_V2; /* If you wonder why we mask this with | |
2627 | * CGROUP_MASK_V2: The 'supported' mask | |
2628 | * might contain pure-V1 or BPF | |
2629 | * controllers, and we never want to | |
2630 | * claim that we could enable those with | |
2631 | * cgroup.subtree_control */ | |
efdb0237 | 2632 | return 0; |
27adcc97 | 2633 | } |
efdb0237 LP |
2634 | |
2635 | r = cg_get_path(SYSTEMD_CGROUP_CONTROLLER, p, "cgroup.subtree_control", &fs); | |
2636 | if (r < 0) | |
2637 | return r; | |
2638 | ||
2639 | for (c = 0; c < _CGROUP_CONTROLLER_MAX; c++) { | |
2640 | CGroupMask bit = CGROUP_CONTROLLER_TO_MASK(c); | |
2641 | const char *n; | |
2642 | ||
ab275f23 LP |
2643 | if (!FLAGS_SET(CGROUP_MASK_V2, bit)) |
2644 | continue; | |
2645 | ||
f99850a0 | 2646 | if (!FLAGS_SET(supported, bit)) |
efdb0237 LP |
2647 | continue; |
2648 | ||
2649 | n = cgroup_controller_to_string(c); | |
2650 | { | |
2651 | char s[1 + strlen(n) + 1]; | |
2652 | ||
f99850a0 | 2653 | s[0] = FLAGS_SET(mask, bit) ? '+' : '-'; |
efdb0237 LP |
2654 | strcpy(s + 1, n); |
2655 | ||
77fa610b LP |
2656 | if (!f) { |
2657 | f = fopen(fs, "we"); | |
54b5ba1d LP |
2658 | if (!f) |
2659 | return log_debug_errno(errno, "Failed to open cgroup.subtree_control file of %s: %m", p); | |
77fa610b LP |
2660 | } |
2661 | ||
604028de | 2662 | r = write_string_stream(f, s, WRITE_STRING_FILE_DISABLE_BUFFER); |
96aa6591 | 2663 | if (r < 0) { |
94f344fb LP |
2664 | log_debug_errno(r, "Failed to %s controller %s for %s (%s): %m", |
2665 | FLAGS_SET(mask, bit) ? "enable" : "disable", n, p, fs); | |
96aa6591 | 2666 | clearerr(f); |
27adcc97 LP |
2667 | |
2668 | /* If we can't turn off a controller, leave it on in the reported resulting mask. This | |
2669 | * happens for example when we attempt to turn off a controller up in the tree that is | |
2670 | * used down in the tree. */ | |
2671 | if (!FLAGS_SET(mask, bit) && r == -EBUSY) /* You might wonder why we check for EBUSY | |
2672 | * only here, and not follow the same logic | |
2673 | * for other errors such as EINVAL or | |
2674 | * EOPNOTSUPP or anything else. That's | |
2675 | * because EBUSY indicates that the | |
2676 | * controllers is currently enabled and | |
2677 | * cannot be disabled because something down | |
2678 | * the hierarchy is still using it. Any other | |
2679 | * error most likely means something like "I | |
2680 | * never heard of this controller" or | |
2681 | * similar. In the former case it's hence | |
2682 | * safe to assume the controller is still on | |
2683 | * after the failed operation, while in the | |
2684 | * latter case it's safer to assume the | |
2685 | * controller is unknown and hence certainly | |
2686 | * not enabled. */ | |
2687 | ret |= bit; | |
2688 | } else { | |
2689 | /* Otherwise, if we managed to turn on a controller, set the bit reflecting that. */ | |
2690 | if (FLAGS_SET(mask, bit)) | |
2691 | ret |= bit; | |
96aa6591 | 2692 | } |
efdb0237 LP |
2693 | } |
2694 | } | |
2695 | ||
27adcc97 LP |
2696 | /* Let's return the precise set of controllers now enabled for the cgroup. */ |
2697 | if (ret_result_mask) | |
2698 | *ret_result_mask = ret; | |
2699 | ||
efdb0237 LP |
2700 | return 0; |
2701 | } | |
2702 | ||
2703 | bool cg_is_unified_wanted(void) { | |
2704 | static thread_local int wanted = -1; | |
415fc41c | 2705 | int r; |
1d84ad94 | 2706 | bool b; |
77fab2a9 | 2707 | const bool is_default = DEFAULT_HIERARCHY == CGROUP_UNIFIED_ALL; |
5f086dc7 | 2708 | _cleanup_free_ char *c = NULL; |
efdb0237 | 2709 | |
77fab2a9 | 2710 | /* If we have a cached value, return that. */ |
efdb0237 LP |
2711 | if (wanted >= 0) |
2712 | return wanted; | |
2713 | ||
239a3d09 ZJS |
2714 | /* If the hierarchy is already mounted, then follow whatever |
2715 | * was chosen for it. */ | |
2716 | if (cg_unified_flush() >= 0) | |
b4cccbc1 | 2717 | return (wanted = unified_cache >= CGROUP_UNIFIED_ALL); |
239a3d09 | 2718 | |
5f086dc7 CD |
2719 | /* If we were explicitly passed systemd.unified_cgroup_hierarchy, |
2720 | * respect that. */ | |
1d84ad94 | 2721 | r = proc_cmdline_get_bool("systemd.unified_cgroup_hierarchy", &b); |
5f086dc7 CD |
2722 | if (r > 0) |
2723 | return (wanted = b); | |
2724 | ||
2725 | /* If we passed cgroup_no_v1=all with no other instructions, it seems | |
2726 | * highly unlikely that we want to use hybrid or legacy hierarchy. */ | |
2727 | r = proc_cmdline_get_key("cgroup_no_v1", 0, &c); | |
2728 | if (r > 0 && streq_ptr(c, "all")) | |
2729 | return (wanted = true); | |
efdb0237 | 2730 | |
5f086dc7 | 2731 | return (wanted = is_default); |
efdb0237 LP |
2732 | } |
2733 | ||
2734 | bool cg_is_legacy_wanted(void) { | |
239a3d09 ZJS |
2735 | static thread_local int wanted = -1; |
2736 | ||
2737 | /* If we have a cached value, return that. */ | |
2738 | if (wanted >= 0) | |
2739 | return wanted; | |
2740 | ||
4e1dfa45 | 2741 | /* Check if we have cgroup v2 already mounted. */ |
1b59cf04 ZJS |
2742 | if (cg_unified_flush() >= 0 && |
2743 | unified_cache == CGROUP_UNIFIED_ALL) | |
239a3d09 | 2744 | return (wanted = false); |
1b59cf04 ZJS |
2745 | |
2746 | /* Otherwise, assume that at least partial legacy is wanted, | |
4e1dfa45 | 2747 | * since cgroup v2 should already be mounted at this point. */ |
239a3d09 | 2748 | return (wanted = true); |
efdb0237 LP |
2749 | } |
2750 | ||
a4464b95 | 2751 | bool cg_is_hybrid_wanted(void) { |
5da38d07 | 2752 | static thread_local int wanted = -1; |
415fc41c | 2753 | int r; |
1d84ad94 | 2754 | bool b; |
c19739db ZJS |
2755 | const bool is_default = DEFAULT_HIERARCHY >= CGROUP_UNIFIED_SYSTEMD; |
2756 | /* We default to true if the default is "hybrid", obviously, | |
2757 | * but also when the default is "unified", because if we get | |
2758 | * called, it means that unified hierarchy was not mounted. */ | |
5da38d07 | 2759 | |
77fab2a9 | 2760 | /* If we have a cached value, return that. */ |
5da38d07 TH |
2761 | if (wanted >= 0) |
2762 | return wanted; | |
2763 | ||
239a3d09 ZJS |
2764 | /* If the hierarchy is already mounted, then follow whatever |
2765 | * was chosen for it. */ | |
2766 | if (cg_unified_flush() >= 0 && | |
2767 | unified_cache == CGROUP_UNIFIED_ALL) | |
2768 | return (wanted = false); | |
2769 | ||
77fab2a9 ZJS |
2770 | /* Otherwise, let's see what the kernel command line has to say. |
2771 | * Since checking is expensive, cache a non-error result. */ | |
1d84ad94 | 2772 | r = proc_cmdline_get_bool("systemd.legacy_systemd_cgroup_controller", &b); |
5da38d07 | 2773 | |
2dcb526d ZJS |
2774 | /* The meaning of the kernel option is reversed wrt. to the return value |
2775 | * of this function, hence the negation. */ | |
77fab2a9 | 2776 | return (wanted = r > 0 ? !b : is_default); |
5da38d07 TH |
2777 | } |
2778 | ||
13c31542 TH |
2779 | int cg_weight_parse(const char *s, uint64_t *ret) { |
2780 | uint64_t u; | |
2781 | int r; | |
2782 | ||
2783 | if (isempty(s)) { | |
2784 | *ret = CGROUP_WEIGHT_INVALID; | |
2785 | return 0; | |
2786 | } | |
2787 | ||
2788 | r = safe_atou64(s, &u); | |
2789 | if (r < 0) | |
2790 | return r; | |
2791 | ||
2792 | if (u < CGROUP_WEIGHT_MIN || u > CGROUP_WEIGHT_MAX) | |
2793 | return -ERANGE; | |
2794 | ||
2795 | *ret = u; | |
2796 | return 0; | |
2797 | } | |
2798 | ||
9be57249 TH |
2799 | const uint64_t cgroup_io_limit_defaults[_CGROUP_IO_LIMIT_TYPE_MAX] = { |
2800 | [CGROUP_IO_RBPS_MAX] = CGROUP_LIMIT_MAX, | |
2801 | [CGROUP_IO_WBPS_MAX] = CGROUP_LIMIT_MAX, | |
ac06a0cf TH |
2802 | [CGROUP_IO_RIOPS_MAX] = CGROUP_LIMIT_MAX, |
2803 | [CGROUP_IO_WIOPS_MAX] = CGROUP_LIMIT_MAX, | |
9be57249 TH |
2804 | }; |
2805 | ||
2806 | static const char* const cgroup_io_limit_type_table[_CGROUP_IO_LIMIT_TYPE_MAX] = { | |
2807 | [CGROUP_IO_RBPS_MAX] = "IOReadBandwidthMax", | |
2808 | [CGROUP_IO_WBPS_MAX] = "IOWriteBandwidthMax", | |
ac06a0cf TH |
2809 | [CGROUP_IO_RIOPS_MAX] = "IOReadIOPSMax", |
2810 | [CGROUP_IO_WIOPS_MAX] = "IOWriteIOPSMax", | |
9be57249 TH |
2811 | }; |
2812 | ||
2813 | DEFINE_STRING_TABLE_LOOKUP(cgroup_io_limit_type, CGroupIOLimitType); | |
2814 | ||
d53d9474 LP |
2815 | int cg_cpu_shares_parse(const char *s, uint64_t *ret) { |
2816 | uint64_t u; | |
2817 | int r; | |
2818 | ||
2819 | if (isempty(s)) { | |
2820 | *ret = CGROUP_CPU_SHARES_INVALID; | |
2821 | return 0; | |
2822 | } | |
2823 | ||
2824 | r = safe_atou64(s, &u); | |
2825 | if (r < 0) | |
2826 | return r; | |
2827 | ||
2828 | if (u < CGROUP_CPU_SHARES_MIN || u > CGROUP_CPU_SHARES_MAX) | |
2829 | return -ERANGE; | |
2830 | ||
2831 | *ret = u; | |
2832 | return 0; | |
2833 | } | |
2834 | ||
2835 | int cg_blkio_weight_parse(const char *s, uint64_t *ret) { | |
2836 | uint64_t u; | |
2837 | int r; | |
2838 | ||
2839 | if (isempty(s)) { | |
2840 | *ret = CGROUP_BLKIO_WEIGHT_INVALID; | |
2841 | return 0; | |
2842 | } | |
2843 | ||
2844 | r = safe_atou64(s, &u); | |
2845 | if (r < 0) | |
2846 | return r; | |
2847 | ||
2848 | if (u < CGROUP_BLKIO_WEIGHT_MIN || u > CGROUP_BLKIO_WEIGHT_MAX) | |
2849 | return -ERANGE; | |
2850 | ||
2851 | *ret = u; | |
2852 | return 0; | |
2853 | } | |
2854 | ||
f0bef277 EV |
2855 | bool is_cgroup_fs(const struct statfs *s) { |
2856 | return is_fs_type(s, CGROUP_SUPER_MAGIC) || | |
2857 | is_fs_type(s, CGROUP2_SUPER_MAGIC); | |
2858 | } | |
2859 | ||
2860 | bool fd_is_cgroup_fs(int fd) { | |
2861 | struct statfs s; | |
2862 | ||
2863 | if (fstatfs(fd, &s) < 0) | |
2864 | return -errno; | |
2865 | ||
2866 | return is_cgroup_fs(&s); | |
2867 | } | |
2868 | ||
b82f71c7 | 2869 | static const char *const cgroup_controller_table[_CGROUP_CONTROLLER_MAX] = { |
efdb0237 LP |
2870 | [CGROUP_CONTROLLER_CPU] = "cpu", |
2871 | [CGROUP_CONTROLLER_CPUACCT] = "cpuacct", | |
13c31542 | 2872 | [CGROUP_CONTROLLER_IO] = "io", |
efdb0237 LP |
2873 | [CGROUP_CONTROLLER_BLKIO] = "blkio", |
2874 | [CGROUP_CONTROLLER_MEMORY] = "memory", | |
3905f127 | 2875 | [CGROUP_CONTROLLER_DEVICES] = "devices", |
03a7b521 | 2876 | [CGROUP_CONTROLLER_PIDS] = "pids", |
17f14955 | 2877 | [CGROUP_CONTROLLER_BPF_FIREWALL] = "bpf-firewall", |
084c7007 | 2878 | [CGROUP_CONTROLLER_BPF_DEVICES] = "bpf-devices", |
efdb0237 LP |
2879 | }; |
2880 | ||
2881 | DEFINE_STRING_TABLE_LOOKUP(cgroup_controller, CGroupController); | |
f98c2585 CD |
2882 | |
2883 | CGroupMask get_cpu_accounting_mask(void) { | |
2884 | static CGroupMask needed_mask = (CGroupMask) -1; | |
2885 | ||
2886 | /* On kernel ≥4.15 with unified hierarchy, cpu.stat's usage_usec is | |
2887 | * provided externally from the CPU controller, which means we don't | |
2888 | * need to enable the CPU controller just to get metrics. This is good, | |
2889 | * because enabling the CPU controller comes at a minor performance | |
2890 | * hit, especially when it's propagated deep into large hierarchies. | |
2891 | * There's also no separate CPU accounting controller available within | |
2892 | * a unified hierarchy. | |
2893 | * | |
2894 | * This combination of factors results in the desired cgroup mask to | |
2895 | * enable for CPU accounting varying as follows: | |
2896 | * | |
2897 | * ╔═════════════════════╤═════════════════════╗ | |
2898 | * ║ Linux ≥4.15 │ Linux <4.15 ║ | |
2899 | * ╔═══════════════╬═════════════════════╪═════════════════════╣ | |
2900 | * ║ Unified ║ nothing │ CGROUP_MASK_CPU ║ | |
2901 | * ╟───────────────╫─────────────────────┼─────────────────────╢ | |
2902 | * ║ Hybrid/Legacy ║ CGROUP_MASK_CPUACCT │ CGROUP_MASK_CPUACCT ║ | |
2903 | * ╚═══════════════╩═════════════════════╧═════════════════════╝ | |
2904 | * | |
2905 | * We check kernel version here instead of manually checking whether | |
2906 | * cpu.stat is present for every cgroup, as that check in itself would | |
2907 | * already be fairly expensive. | |
2908 | * | |
2909 | * Kernels where this patch has been backported will therefore have the | |
2910 | * CPU controller enabled unnecessarily. This is more expensive than | |
2911 | * necessary, but harmless. ☺️ | |
2912 | */ | |
2913 | ||
2914 | if (needed_mask == (CGroupMask) -1) { | |
2915 | if (cg_all_unified()) { | |
2916 | struct utsname u; | |
2917 | assert_se(uname(&u) >= 0); | |
2918 | ||
2919 | if (str_verscmp(u.release, "4.15") < 0) | |
2920 | needed_mask = CGROUP_MASK_CPU; | |
2921 | else | |
2922 | needed_mask = 0; | |
2923 | } else | |
2924 | needed_mask = CGROUP_MASK_CPUACCT; | |
2925 | } | |
2926 | ||
2927 | return needed_mask; | |
2928 | } | |
2929 | ||
2930 | bool cpu_accounting_is_cheap(void) { | |
2931 | return get_cpu_accounting_mask() == 0; | |
2932 | } |